Rhasspy and Home Assistant Voice with Dr. Michael Hansen

there's something new from our sponsor there's something new from our sponsor text control for decades TX text text control for decades TX text text control for decades TX text controls rendering was tied to Windows controls rendering was tied to Windows controls rendering was tied to Windows functionality but now it's a new era TX functionality but now it's a new era TX functionality but now it's a new era TX text control is platform independent you text control is platform independent you text control is platform independent you can run your net applications on Windows can run your net applications on Windows can run your net applications on Windows or Linux deploy with Docker containers or Linux deploy with Docker containers or Linux deploy with Docker containers kubernetes or host directly in Azure app kubernetes or host directly in Azure app kubernetes or host directly in Azure app Services seamlessly integrate document Services seamlessly integrate document Services seamlessly integrate document editing signing collaboration and PDF editing signing collaboration and PDF editing signing collaboration and PDF processing into your asp.net core and processing into your asp.net core and processing into your asp.net core and angular applications are you ready to angular applications are you ready to angular applications are you ready to see it in action see it in action see it in action check out the live demo at demos. text check out the live demo at demos. text check out the live demo at demos. text control.com that's demos. text [Music] [Music] control.com hi I'm Scott hansman this is control.com hi I'm Scott hansman this is control.com hi I'm Scott hansman this is another episode of cancel minutes today another episode of cancel minutes today another episode of cancel minutes today I'm chatting with Dr Michael Hansen he I'm chatting with Dr Michael Hansen he I'm chatting with Dr Michael Hansen he is the primary on the raspy voice assist is the primary on the raspy voice assist is the primary on the raspy voice assist assistant and this is a fully open assistant and this is a fully open assistant and this is a fully open source set of voice assistant Services source set of voice assistant Services source set of voice assistant Services it's super exciting and I'm thrilled to it's super exciting and I'm thrilled to it's super exciting and I'm thrilled to be chatting with you today sir hi Scott be chatting with you today sir hi Scott be chatting with you today sir hi Scott uh thanks for having me it's an honor to uh thanks for having me it's an honor to uh thanks for having me it's an honor to be here yeah so I've had paulus on from be here yeah so I've had paulus on from be here yeah so I've had paulus on from home assistant maybe you could kind of home assistant maybe you could kind of home assistant maybe you could kind of put it all into context what is raspby put it all into context what is raspby put it all into context what is raspby how does it relate to home assistant and how does it relate to home assistant and how does it relate to home assistant and the other things in that in that space the other things in that in that space the other things in that in that space yeah sure so I'm I work for nabasa which yeah sure so I'm I work for nabasa which yeah sure so I'm I work for nabasa which is the company that uh funds the is the company that uh funds the is the company that uh funds the development of Home assistant and so development of Home assistant and so development of Home assistant and so they started back in in 2023 uh a a

they started back in in 2023 uh a a they started back in in 2023 uh a a year-long push for adding voice to to year-long push for adding voice to to year-long push for adding voice to to home assistant and to the the open home home assistant and to the the open home home assistant and to the the open home as they as they call it um so raspy was as they as they call it um so raspy was as they as they call it um so raspy was a a hobby project for me for a number of a a hobby project for me for a number of a a hobby project for me for a number of years and they approached me and said years and they approached me and said years and they approached me and said you know you're this open source voice you know you're this open source voice you know you're this open source voice guy would you like to be you know the guy would you like to be you know the guy would you like to be you know the voice guy for uh for the year of voice voice guy for uh for the year of voice voice guy for uh for the year of voice and I said sure and so that's where we and I said sure and so that's where we and I said sure and so that's where we are today we've been pushing to get uh are today we've been pushing to get uh are today we've been pushing to get uh open source comp for voice into home open source comp for voice into home open source comp for voice into home assistant so home assistant exists for a assistant so home assistant exists for a assistant so home assistant exists for a number of reasons but one could argue number of reasons but one could argue number of reasons but one could argue that it exists as a void a vote of no that it exists as a void a vote of no that it exists as a void a vote of no confidence against some of the confidence against some of the confidence against some of the cloud-based assistance whether it be the cloud-based assistance whether it be the cloud-based assistance whether it be the home stuff that's built into apple or home stuff that's built into apple or home stuff that's built into apple or whether it's Alexa or things like that whether it's Alexa or things like that whether it's Alexa or things like that um does raspy have a similar philosophy um does raspy have a similar philosophy um does raspy have a similar philosophy or do you am I mischaracterizing that I or do you am I mischaracterizing that I or do you am I mischaracterizing that I I would vote it not not not say it's a I would vote it not not not say it's a I would vote it not not not say it's a vote of no confidence but more of a uh a vote of no confidence but more of a uh a vote of no confidence but more of a uh a desire to keep Choice available to desire to keep Choice available to desire to keep Choice available to people H so people can still hook up people H so people can still hook up people H so people can still hook up cloud services to home assistant if they cloud services to home assistant if they cloud services to home assistant if they choose it's just that they they need to choose it's just that they they need to choose it's just that they they need to make that choice and not have it forced make that choice and not have it forced make that choice and not have it forced upon them and uh raspby is the same way upon them and uh raspby is the same way upon them and uh raspby is the same way uh there it's it's got modular uh there it's it's got modular uh there it's it's got modular components so you could hook up you know components so you could hook up you know components so you could hook up you know a speech detect system that that went a speech detect system that that went a speech detect system that that went out to the cloud if you wanted to and a out to the cloud if you wanted to and a out to the cloud if you wanted to and a number of people did because it's just number of people did because it's just number of people did because it's just hard it's hard to beat a data center hard it's hard to beat a data center hard it's hard to beat a data center right it refers as a Raspberry Pi um but right it refers as a Raspberry Pi um but right it refers as a Raspberry Pi um but it was kind of a local first and then it was kind of a local first and then it was kind of a local first and then choose if you you want to put pieces out choose if you you want to put pieces out choose if you you want to put pieces out there online on the cloud last time I

there online on the cloud last time I there online on the cloud last time I had paulus on he was had paulus on he was had paulus on he was 2021 show 2021 show 2021 show 788 and this show will be somewhere in 788 and this show will be somewhere in 788 and this show will be somewhere in the nine in the 900s so that's about the nine in the 900s so that's about the nine in the 900s so that's about four years it was in the thick of uh of four years it was in the thick of uh of four years it was in the thick of uh of covid and the thing that I had told him covid and the thing that I had told him covid and the thing that I had told him was that I think that Alexa had been was that I think that Alexa had been was that I think that Alexa had been keeping me on that ecosystem it was just keeping me on that ecosystem it was just keeping me on that ecosystem it was just the simple the simple idea that I could the simple the simple idea that I could the simple the simple idea that I could just holler and the light would go on or just holler and the light would go on or just holler and the light would go on or off or I could automate things and at off or I could automate things and at off or I could automate things and at the time in 2021 home assistant was very the time in 2021 home assistant was very the time in 2021 home assistant was very touchscreen focused people would build touchscreen focused people would build touchscreen focused people would build Central touch screens in the center of Central touch screens in the center of Central touch screens in the center of their house and that's how they would do their house and that's how they would do their house and that's how they would do their things or they do it with their their things or they do it with their their things or they do it with their with their phones why did it take so with their phones why did it take so with their phones why did it take so long for for audio to happen in that long for for audio to happen in that long for for audio to happen in that ecosystem I mean they were looking for ecosystem I mean they were looking for ecosystem I mean they were looking for the right people and for waiting for the the right people and for waiting for the the right people and for waiting for the technology to get there and I to me the technology to get there and I to me the technology to get there and I to me the big change happened uh when they got ESP big change happened uh when they got ESP big change happened uh when they got ESP home working really well so uh ESP home home working really well so uh ESP home home working really well so uh ESP home is a software that is is part of the is a software that is is part of the is a software that is is part of the home assistant project it runs on home assistant project it runs on home assistant project it runs on firmware on these small little ESP chips firmware on these small little ESP chips firmware on these small little ESP chips and you can I I have I have them running and you can I I have I have them running and you can I I have I have them running all over my house actually I got a all over my house actually I got a all over my house actually I got a sensor board in the breaker box that's sensor board in the breaker box that's sensor board in the breaker box that's monitoring whole home electricity I've monitoring whole home electricity I've monitoring whole home electricity I've got one in the garage door that uh got one in the garage door that uh got one in the garage door that uh triggers the garage door to open and triggers the garage door to open and triggers the garage door to open and senses whether or not it's open with a senses whether or not it's open with a senses whether or not it's open with a magnet and a read switch and so what magnet and a read switch and so what magnet and a read switch and so what what this firmware does is is connects what this firmware does is is connects what this firmware does is is connects the home assistant and kind of uh the home assistant and kind of uh the home assistant and kind of uh presents presents it with a view that um presents presents it with a view that um presents presents it with a view that um in your home assistant you can see oh I in your home assistant you can see oh I in your home assistant you can see oh I have a sensor I have a switch you know

have a sensor I have a switch you know have a sensor I have a switch you know something just like a light switch or something just like a light switch or something just like a light switch or something like that so it acts as a nice something like that so it acts as a nice something like that so it acts as a nice uh abstract layer over the firmware and uh abstract layer over the firmware and uh abstract layer over the firmware and over the over the over the device and with ESP home in place it device and with ESP home in place it device and with ESP home in place it became pretty clear that that was going became pretty clear that that was going became pretty clear that that was going to be the path to get voice to so extend to be the path to get voice to so extend to be the path to get voice to so extend espb home with voice espb home with voice espb home with voice capability and that was going to be how capability and that was going to be how capability and that was going to be how we' get a device and that's what we we' get a device and that's what we we' get a device and that's what we ended up doing okay so I know that it's ended up doing okay so I know that it's ended up doing okay so I know that it's I I'm the first person to say never say I I'm the first person to say never say I I'm the first person to say never say just when describing hard work you know just when describing hard work you know just when describing hard work you know like oh why didn't you just why didn't like oh why didn't you just why didn't like oh why didn't you just why didn't you just but as a as a rhetorical you just but as a as a rhetorical you just but as a as a rhetorical argument as a rhetorical moment to like argument as a rhetorical moment to like argument as a rhetorical moment to like uh move the conversation forward why was uh move the conversation forward why was uh move the conversation forward why was this hard why why is it just not a this hard why why is it just not a this hard why why is it just not a listener into some cloud service or why listener into some cloud service or why listener into some cloud service or why isn't it just record something and then isn't it just record something and then isn't it just record something and then throw it at the cloud sure weren't there throw it at the cloud sure weren't there throw it at the cloud sure weren't there simpler ways to do this there there were simpler ways to do this there there were simpler ways to do this there there were uh I mean in some sense it it does do uh I mean in some sense it it does do uh I mean in some sense it it does do that in the the sense that it it just that in the the sense that it it just that in the the sense that it it just records and throws it to your home records and throws it to your home records and throws it to your home assistant assistant assistant server um but the big uh hurdle was server um but the big uh hurdle was server um but the big uh hurdle was getting everything such that it could be getting everything such that it could be getting everything such that it could be open source and you could have choice open source and you could have choice open source and you could have choice for local present and so we we did for local present and so we we did for local present and so we we did achieve that so the the we we sell a achieve that so the the we we sell a achieve that so the the we we sell a device called the home assistant preview device called the home assistant preview device called the home assistant preview Edition uh it's it's an ESP device that Edition uh it's it's an ESP device that Edition uh it's it's an ESP device that runs open source firmware with espb home runs open source firmware with espb home runs open source firmware with espb home and it also has a little chip on there and it also has a little chip on there and it also has a little chip on there an audio chip for from a company called an audio chip for from a company called an audio chip for from a company called xmos or xmos or xmos or xmos uh you'll find them in speaker xmos uh you'll find them in speaker xmos uh you'll find them in speaker phones and stuff but it it does audio phones and stuff but it it does audio phones and stuff but it it does audio cleanup and a lot of a lot of nice

cleanup and a lot of a lot of nice cleanup and a lot of a lot of nice things uh with with audio can Echo things uh with with audio can Echo things uh with with audio can Echo cancellation Etc uh but it now it also cancellation Etc uh but it now it also cancellation Etc uh but it now it also runs open source firmware and so we're runs open source firmware and so we're runs open source firmware and so we're we're trying to get the whole stack here we're trying to get the whole stack here we're trying to get the whole stack here to be hackable open source and free for to be hackable open source and free for to be hackable open source and free for people so so that they can uh I I guess people so so that they can uh I I guess people so so that they can uh I I guess they can have confidence and they can they can have confidence and they can they can have confidence and they can also change things as they also change things as they also change things as they please do all the home assist please do all the home assist please do all the home assist and I say that with like the the generic and I say that with like the the generic and I say that with like the the generic noun do all the different things whether noun do all the different things whether noun do all the different things whether it be Siri or Alexa you know do they all it be Siri or Alexa you know do they all it be Siri or Alexa you know do they all send stuff up into the cloud to send stuff up into the cloud to send stuff up into the cloud to recognize our voices is there any voice recognize our voices is there any voice recognize our voices is there any voice recognition other than the Wake word recognition other than the Wake word recognition other than the Wake word happening locally on the kind of the happening locally on the kind of the happening locally on the kind of the traditional voice assistants that are traditional voice assistants that are traditional voice assistants that are available not as far as I know and I available not as far as I know and I available not as far as I know and I believe at least in Amazon's case the believe at least in Amazon's case the believe at least in Amazon's case the Wake word also uh there's a there's Wake word also uh there's a there's Wake word also uh there's a there's there's two models I believe there's a there's two models I believe there's a there's two models I believe there's a the local model that triggers first and the local model that triggers first and the local model that triggers first and then there's a a check with the Cloud to then there's a a check with the Cloud to then there's a a check with the Cloud to to double check that it actually to double check that it actually to double check that it actually happened or that you actually said the happened or that you actually said the happened or that you actually said the word which we we may end up adopting a word which we we may end up adopting a word which we we may end up adopting a similar approach except having the similar approach except having the similar approach except having the larger um model run on like the home larger um model run on like the home larger um model run on like the home assistant server because it's it's assistant server because it's it's assistant server because it's it's really difficult to to get some of these really difficult to to get some of these really difficult to to get some of these things to run on such a tiny tiny chip I things to run on such a tiny tiny chip I things to run on such a tiny tiny chip I mean these things are in the the 200 mean these things are in the the 200 mean these things are in the the 200 megahertz range which you know when I megahertz range which you know when I megahertz range which you know when I was growing up that was a big deal like was growing up that was a big deal like was growing up that was a big deal like that was fast right but uh nowadays it's that was fast right but uh nowadays it's that was fast right but uh nowadays it's you can barely run anything it seems on you can barely run anything it seems on you can barely run anything it seems on 200 MHz and and 16 Megs of RAM 200 MHz and and 16 Megs of RAM 200 MHz and and 16 Megs of RAM I remember doing really really really I remember doing really really really I remember doing really really really basic uh voice to voice to text you know

basic uh voice to voice to text you know basic uh voice to voice to text you know Dragon Naturally Speaking of course Dragon Naturally Speaking of course Dragon Naturally Speaking of course running it on running it on running it on 46s uh pennium uh they were very 46s uh pennium uh they were very 46s uh pennium uh they were very constrained rather than allowing someone constrained rather than allowing someone constrained rather than allowing someone to say anything some of the early voice to say anything some of the early voice to say anything some of the early voice systems were like here's 20 words that systems were like here's 20 words that systems were like here's 20 words that it knows yeah and if you dug into it it it knows yeah and if you dug into it it it knows yeah and if you dug into it it wasn't that it really knew the word it wasn't that it really knew the word it wasn't that it really knew the word it was either comparing waveforms or it was was either comparing waveforms or it was was either comparing waveforms or it was listening for hard conidence M you know listening for hard conidence M you know listening for hard conidence M you know uh can can you talk about the difference uh can can you talk about the difference uh can can you talk about the difference between how early voice recognition between how early voice recognition between how early voice recognition worked and then what Chang that allow us worked and then what Chang that allow us worked and then what Chang that allow us to ask open-ended questions with with to ask open-ended questions with with to ask open-ended questions with with voice well and actually I'll I'll voice well and actually I'll I'll voice well and actually I'll I'll contrast the open-ended too with what contrast the open-ended too with what contrast the open-ended too with what raspy does because I think there's an raspy does because I think there's an raspy does because I think there's an important difference important difference important difference there um so you know early on yeah there there um so you know early on yeah there there um so you know early on yeah there was uh usually it was it was like was uh usually it was it was like was uh usually it was it was like looking for hard-coded looking for hard-coded looking for hard-coded waveforms um or or like you said waveforms um or or like you said waveforms um or or like you said consonants but but basically uh spectrog consonants but but basically uh spectrog consonants but but basically uh spectrog graphical features right in in the graphical features right in in the graphical features right in in the spectrogram features mhm it still does spectrogram features mhm it still does spectrogram features mhm it still does that except the method for doing it is that except the method for doing it is that except the method for doing it is is much more sophisticated I mean is much more sophisticated I mean is much more sophisticated I mean nowadays using neural networks and stuff nowadays using neural networks and stuff nowadays using neural networks and stuff there's there's all sorts of of ways to there's there's all sorts of of ways to there's there's all sorts of of ways to do it uh with that uh what so that's for do it uh with that uh what so that's for do it uh with that uh what so that's for for open-ended you you these days you for open-ended you you these days you for open-ended you you these days you would train some neural network and you would train some neural network and you would train some neural network and you know be Off to the Races but it it I know be Off to the Races but it it I know be Off to the Races but it it I don't know you need huge ones to have don't know you need huge ones to have don't know you need huge ones to have very open-ended questions and this is very open-ended questions and this is very open-ended questions and this is where we run into problems when trying where we run into problems when trying where we run into problems when trying to run things to run things to run things locally um the where I kind of settled locally um the where I kind of settled locally um the where I kind of settled with raspy there's a speech to text with raspy there's a speech to text with raspy there's a speech to text program called program called program called CDI uh that kind of had a had a really CDI uh that kind of had a had a really CDI uh that kind of had a had a really nice middle ground it it wasn't fully

nice middle ground it it wasn't fully nice middle ground it it wasn't fully neural network based but it also so you neural network based but it also so you neural network based but it also so you could ask open-ended questions but there could ask open-ended questions but there could ask open-ended questions but there was still some constraint you had to was still some constraint you had to was still some constraint you had to give it lots of text and it could it give it lots of text and it could it give it lots of text and it could it could answer this answer certain could answer this answer certain could answer this answer certain questions or it could you know recognize questions or it could you know recognize questions or it could you know recognize certain certain certain questions um and I found that the questions um and I found that the questions um and I found that the open-ended stuff was performed really open-ended stuff was performed really open-ended stuff was performed really poorly with with these things um and so poorly with with these things um and so poorly with with these things um and so the approach I took with raspby was to the approach I took with raspby was to the approach I took with raspby was to kind of change the question from what kind of change the question from what kind of change the question from what did the user say to which of these did the user say to which of these did the user say to which of these things did the user things did the user things did the user say and that was a a major difference uh say and that was a a major difference uh say and that was a a major difference uh in terms of accuracy and everything in terms of accuracy and everything in terms of accuracy and everything because if you're if you're saying Turn because if you're if you're saying Turn because if you're if you're saying Turn On The Lights to your voice assistant On The Lights to your voice assistant On The Lights to your voice assistant you probably also don't need to say you you probably also don't need to say you you probably also don't need to say you know Grandma baking cookies you know or know Grandma baking cookies you know or know Grandma baking cookies you know or some other sentence you don't need the some other sentence you don't need the some other sentence you don't need the whole the whole English language at your whole the whole English language at your whole the whole English language at your finger or at your voice right do you finger or at your voice right do you finger or at your voice right do you then ignore it if they say extraneous then ignore it if they say extraneous then ignore it if they say extraneous things so that was a a tricky thing that things so that was a a tricky thing that things so that was a a tricky thing that I've only recently got to work robustly I've only recently got to work robustly I've only recently got to work robustly um adding an option that says it's not um adding an option that says it's not um adding an option that says it's not any of those things um CDI unless you go any of those things um CDI unless you go any of those things um CDI unless you go a certain route it it it will try as a certain route it it it will try as a certain route it it it will try as hard as as computer possible to uh fit hard as as computer possible to uh fit hard as as computer possible to uh fit what you said to one of those phrases what you said to one of those phrases what you said to one of those phrases and that was a problem too because um and that was a problem too because um and that was a problem too because um especially uh in the early early days especially uh in the early early days especially uh in the early early days wake words were so poor that you get wake words were so poor that you get wake words were so poor that you get false activations quite a bit and if you false activations quite a bit and if you false activations quite a bit and if you combine that with a a kind of a closed combine that with a a kind of a closed combine that with a a kind of a closed system where it was uh you know

system where it was uh you know system where it was uh you know recognize what just from this set of recognize what just from this set of recognize what just from this set of phrases that you could uh that you could phrases that you could uh that you could phrases that you could uh that you could say then you would almost surely get say then you would almost surely get say then you would almost surely get some random light in your house to turn some random light in your house to turn some random light in your house to turn on or or something you know on or or something you know on or or something you know unended what is the difference so CDI is unended what is the difference so CDI is unended what is the difference so CDI is spelled kld D and it's the CDI ASR spelled kld D and it's the CDI ASR spelled kld D and it's the CDI ASR project what is and they've been around project what is and they've been around project what is and they've been around for a decade uh and really really active for a decade uh and really really active for a decade uh and really really active in the last four or five years can you in the last four or five years can you in the last four or five years can you explain the difference in explain the difference in explain the difference in responsibilities from a computer science responsibilities from a computer science responsibilities from a computer science perspective between what they do and perspective between what they do and perspective between what they do and then what your piece does and what what then what your piece does and what what then what your piece does and what what you're allowed to say and what you're you're allowed to say and what you're you're allowed to say and what you're not allowed to say because I'm not quite not allowed to say because I'm not quite not allowed to say because I'm not quite understanding what CDI does and what understanding what CDI does and what understanding what CDI does and what raspy does ah so raspy is kind of a a raspy does ah so raspy is kind of a a raspy does ah so raspy is kind of a a toolkit where you can plug and play toolkit where you can plug and play toolkit where you can plug and play different speech to text text to speech different speech to text text to speech different speech to text text to speech systems and build like build your own systems and build like build your own systems and build like build your own voice assistant uh by putting components voice assistant uh by putting components voice assistant uh by putting components together one of those components is CDI together one of those components is CDI together one of those components is CDI and I used CDI in a very particular way and I used CDI in a very particular way and I used CDI in a very particular way where you could come in and say these where you could come in and say these where you could come in and say these are the voice commands I'd like you to are the voice commands I'd like you to are the voice commands I'd like you to recognize and you would tell up front recognize and you would tell up front recognize and you would tell up front and then it would program CDI to do that and then it would program CDI to do that and then it would program CDI to do that CDI can be used in a more open-ended way CDI can be used in a more open-ended way CDI can be used in a more open-ended way as well I see and I know that open as well I see and I know that open as well I see and I know that open source has been like fundamentally source has been like fundamentally source has been like fundamentally important to everything that home important to everything that home important to everything that home assistant does and presumably as well assistant does and presumably as well assistant does and presumably as well all the way to raspy but then I all the way to raspy but then I all the way to raspy but then I understand that if you start using uh understand that if you start using uh understand that if you start using uh you know models that have been trained you know models that have been trained you know models that have been trained that it gets ethically squishy around that it gets ethically squishy around that it gets ethically squishy around what the data was trained on CDI what the data was trained on CDI what the data was trained on CDI includes you know two dozen models that includes you know two dozen models that includes you know two dozen models that you can get in different languages and you can get in different languages and you can get in different languages and there's uh there's English there's there's uh there's English there's there's uh there's English there's multiple English ones there's a manderin multiple English ones there's a manderin multiple English ones there's a manderin there's there's Arabic do we need to

there's there's Arabic do we need to there's there's Arabic do we need to think about that if we're really focused think about that if we're really focused think about that if we're really focused on an open source home assistant do we on an open source home assistant do we on an open source home assistant do we have to think about the models and where have to think about the models and where have to think about the models and where they came from or does the fact that they came from or does the fact that they came from or does the fact that they're pluggable make it not that big they're pluggable make it not that big they're pluggable make it not that big of a deal I mean it it it lets it pushes of a deal I mean it it it lets it pushes of a deal I mean it it it lets it pushes the responsibility onto the user so you the responsibility onto the user so you the responsibility onto the user so you need to make sure that you're in need to make sure that you're in need to make sure that you're in compliance with whatever the license is compliance with whatever the license is compliance with whatever the license is if you can plug in your whatever model if you can plug in your whatever model if you can plug in your whatever model um so I I took that to heart though and um so I I took that to heart though and um so I I took that to heart though and as part of the raspby project I've as part of the raspby project I've as part of the raspby project I've trained a number of CDI models off of trained a number of CDI models off of trained a number of CDI models off of mozilla's common voice data set so their mozilla's common voice data set so their mozilla's common voice data set so their stuff is all public domain uh and they stuff is all public domain uh and they stuff is all public domain uh and they they work hard to source that you know they work hard to source that you know they work hard to source that you know correctly and uh and according to a correctly and uh and according to a correctly and uh and according to a public domain license and data is really public domain license and data is really public domain license and data is really where things have been tough um where things have been tough um where things have been tough um especially for languages outside of you especially for languages outside of you especially for languages outside of you know English German French Spanish you know English German French Spanish you know English German French Spanish you know a couple a couple of big ones know a couple a couple of big ones know a couple a couple of big ones and so I've uh as part of part of raspy and so I've uh as part of part of raspy and so I've uh as part of part of raspy and really more now as part of and really more now as part of and really more now as part of naasa um I reach out to people for for naasa um I reach out to people for for naasa um I reach out to people for for you know languages that are unusual for you know languages that are unusual for you know languages that are unusual for a voice assistant and try to try to see a voice assistant and try to try to see a voice assistant and try to try to see if I can nudge them to record for if I can nudge them to record for if I can nudge them to record for us the some of the models at CDI are us the some of the models at CDI are us the some of the models at CDI are gosh 8 n years old uh and they're still gosh 8 n years old uh and they're still gosh 8 n years old uh and they're still you know a gigabyte half gig how much in you know a gigabyte half gig how much in you know a gigabyte half gig how much in order to run a home assistant on a order to run a home assistant on a order to run a home assistant on a constrained piece of Hardware what is a constrained piece of Hardware what is a constrained piece of Hardware what is a correct model for what's the correct correct model for what's the correct correct model for what's the correct size and and and what's the like is size and and and what's the like is size and and and what's the like is there an 8020 rule that applies to voice there an 8020 rule that applies to voice there an 8020 rule that applies to voice models as they you know get like you models as they you know get like you models as they you know get like you want a low a low error rate but you

want a low a low error rate but you want a low a low error rate but you don't want to download 10 gigs of some don't want to download 10 gigs of some don't want to download 10 gigs of some blob in order to do it so that's part of blob in order to do it so that's part of blob in order to do it so that's part of where I saved a lot of space the where I saved a lot of space the where I saved a lot of space the gigabyte that you're seeing for most of gigabyte that you're seeing for most of gigabyte that you're seeing for most of those models is the language model those models is the language model those models is the language model that's trained on like gigatex or that's trained on like gigatex or that's trained on like gigatex or something to recognize uh it has something to recognize uh it has something to recognize uh it has statistics of which words follows others statistics of which words follows others statistics of which words follows others words and I don't need that in in the words and I don't need that in in the words and I don't need that in in the case where you program your own voice case where you program your own voice case where you program your own voice commands it it produces a bespoke commands it it produces a bespoke commands it it produces a bespoke language model that just covers the language model that just covers the language model that just covers the space that you've given it which is is space that you've given it which is is space that you've given it which is is is not the entire internet right so it's is not the entire internet right so it's is not the entire internet right so it's ends up being a few hundred kilobytes or ends up being a few hundred kilobytes or ends up being a few hundred kilobytes or a few megabytes at worst then is it fair a few megabytes at worst then is it fair a few megabytes at worst then is it fair to say that that would then make it to say that that would then make it to say that that would then make it really fast like the fact that it's so really fast like the fact that it's so really fast like the fact that it's so small like you know there's there's small like you know there's there's small like you know there's there's always that pregnant pause after you ask always that pregnant pause after you ask always that pregnant pause after you ask Alexa to do something right so people Alexa to do something right so people Alexa to do something right so people are often quite amazed when at how at are often quite amazed when at how at are often quite amazed when at how at the speed I mean the speed I mean the speed I mean uh even on a Raspberry Pi 4 you can uh even on a Raspberry Pi 4 you can uh even on a Raspberry Pi 4 you can usually get subsec responses uh from it usually get subsec responses uh from it usually get subsec responses uh from it and on a on a desktop computer I mean and on a on a desktop computer I mean and on a on a desktop computer I mean it's it's under a tenth of a second it's it's under a tenth of a second it's it's under a tenth of a second easily to wow to get responses back now easily to wow to get responses back now easily to wow to get responses back now there's there is a problem I mean CDI uh there's there is a problem I mean CDI uh there's there is a problem I mean CDI uh I'm using their their neural network I'm using their their neural network I'm using their their neural network models which are very early neural models which are very early neural models which are very early neural networks so they're very they're very networks so they're very they're very networks so they're very they're very fast on Modern Hardware but if you fast on Modern Hardware but if you fast on Modern Hardware but if you provide too many possibilities or ones provide too many possibilities or ones provide too many possibilities or ones that are very close together um like 15 that are very close together um like 15 that are very close together um like 15 and and and 50 you sometimes have to enunciate a 50 you sometimes have to enunciate a 50 you sometimes have to enunciate a little more which can be frustrating for little more which can be frustrating for little more which can be frustrating for people um people um people um so yeah I've I've if you if you pair it so yeah I've I've if you if you pair it so yeah I've I've if you if you pair it down to just what you need then uh then

down to just what you need then uh then down to just what you need then uh then it works very well but if you try to it works very well but if you try to it works very well but if you try to blow it up which with combinatorics it's blow it up which with combinatorics it's blow it up which with combinatorics it's very easy to blow it up uh to millions very easy to blow it up uh to millions very easy to blow it up uh to millions of possible sentences right when you of possible sentences right when you of possible sentences right when you start including start including start including numbers have you ever seen the uh the numbers have you ever seen the uh the numbers have you ever seen the uh the skit where the two Scottish gentlemen skit where the two Scottish gentlemen skit where the two Scottish gentlemen get stuck on an elev that's a voice get stuck on an elev that's a voice get stuck on an elev that's a voice activated elevator no I haven't seen activated elevator no I haven't seen activated elevator no I haven't seen that I would love to it's it's kind of a that I would love to it's it's kind of a that I would love to it's it's kind of a legend but basically these two Scottish legend but basically these two Scottish legend but basically these two Scottish fellas are on an elevator and one of the fellas are on an elevator and one of the fellas are on an elevator and one of the an Englishman is on the elevator with an Englishman is on the elevator with an Englishman is on the elevator with them and he talks about how wonderful it them and he talks about how wonderful it them and he talks about how wonderful it is that this elevator is voice activated is that this elevator is voice activated is that this elevator is voice activated and he goes and he says you know floor and he goes and he says you know floor and he goes and he says you know floor 12 and then he gets off at floor 12 and 12 and then he gets off at floor 12 and 12 and then he gets off at floor 12 and then he leaves and the scotsmen are like then he leaves and the scotsmen are like then he leaves and the scotsmen are like this is amazing and then they go 11 and this is amazing and then they go 11 and this is amazing and then they go 11 and they're just stuck and they they're they're just stuck and they they're they're just stuck and they they're Scream one of them climbs on top of the Scream one of them climbs on top of the Scream one of them climbs on top of the other and they're yelling into the other and they're yelling into the other and they're yelling into the speaker and they just like they can't speaker and they just like they can't speaker and they just like they can't get off until this thing lets them off get off until this thing lets them off get off until this thing lets them off and there's no buttons to push they're and there's no buttons to push they're and there's no buttons to push they're just screaming a perfect so I'm just screaming a perfect so I'm just screaming a perfect so I'm wondering in in mixed accent houses like wondering in in mixed accent houses like wondering in in mixed accent houses like my like my own is that an interesting my like my own is that an interesting my like my own is that an interesting challenge to solve even though we're all challenge to solve even though we're all challenge to solve even though we're all speaking English that is a very speaking English that is a very speaking English that is a very interesting challenge uh We've interesting challenge uh We've interesting challenge uh We've especially had that challenge with our especially had that challenge with our especially had that challenge with our wake wordss um that they we've had to we wake wordss um that they we've had to we wake wordss um that they we've had to we we've we started something called a we've we started something called a we've we started something called a wward wward wward collective um which sounds larger than collective um which sounds larger than collective um which sounds larger than it is but we we're working working on it is but we we're working working on it is but we we're working working on collecting as much data as we can we collecting as much data as we can we collecting as much data as we can we need more data from from people with need more data from from people with need more data from from people with accented English alth um and then and accented English alth um and then and accented English alth um and then and then we need uh um basically more more then we need uh um basically more more then we need uh um basically more more female or high register vocals to to female or high register vocals to to female or high register vocals to to because most of it is is you know like because most of it is is you know like because most of it is is you know like males uh from us English is like just

males uh from us English is like just males uh from us English is like just like lots of data sets right um speech like lots of data sets right um speech like lots of data sets right um speech to text we do we do have the same to text we do we do have the same to text we do we do have the same problems though with with accented problems though with with accented problems though with with accented English there are some models that are English there are some models that are English there are some models that are tuned specifically for particular tuned specifically for particular tuned specifically for particular regions like um Indian English regions like um Indian English regions like um Indian English or others uh but I would love to train a or others uh but I would love to train a or others uh but I would love to train a model um not just for Scottish accented model um not just for Scottish accented model um not just for Scottish accented English but for Scots itself I've heard English but for Scots itself I've heard English but for Scots itself I've heard that's kind of a for for some people that's kind of a for for some people that's kind of a for for some people it's like a white whale situation to get it's like a white whale situation to get it's like a white whale situation to get a model trained um for for a model trained um for for a model trained um for for Scots so can I train help me understand Scots so can I train help me understand Scots so can I train help me understand the relationship between me as a the relationship between me as a the relationship between me as a technical user and you as the expert are technical user and you as the expert are technical user and you as the expert are you training the models or because I'm you training the models or because I'm you training the models or because I'm making the assistant I'm I'm giving it a making the assistant I'm I'm giving it a making the assistant I'm I'm giving it a list of all the things I want it to be list of all the things I want it to be list of all the things I want it to be able to say and I could do custom things able to say and I could do custom things able to say and I could do custom things like water the garden yep which you know like water the garden yep which you know like water the garden yep which you know and things like that but if I've got a and things like that but if I've got a and things like that but if I've got a mixed uh accent house can we train our mixed uh accent house can we train our mixed uh accent house can we train our own models and how many books do I have own models and how many books do I have own models and how many books do I have to read in order for you to to to train to read in order for you to to to train to read in order for you to to to train that how's that work that uh you you that how's that work that uh you you that how's that work that uh you you could technically train your own model could technically train your own model could technically train your own model there is a fine-tuning method that can there is a fine-tuning method that can there is a fine-tuning method that can be used but I've I have yet to ever do be used but I've I have yet to ever do be used but I've I have yet to ever do that what I've done instead is is just that what I've done instead is is just that what I've done instead is is just tried to gather voice data and train a a tried to gather voice data and train a a tried to gather voice data and train a a model for a particular accent um and and model for a particular accent um and and model for a particular accent um and and hope for the best there hope for the best there hope for the best there um what what's the difference here is um what what's the difference here is um what what's the difference here is where I I would be training the acoustic where I I would be training the acoustic where I I would be training the acoustic model this is the model that takes in model this is the model that takes in model this is the model that takes in the audio waveform and produces uh the audio waveform and produces uh the audio waveform and produces uh probabilities of probabilities of probabilities of phones so phones for people who don't phones so phones for people who don't phones so phones for people who don't know this if uh human speech there's

know this if uh human speech there's know this if uh human speech there's there's certain fundamental sounds that there's certain fundamental sounds that there's certain fundamental sounds that you know we we can make as humans and um you know we we can make as humans and um you know we we can make as humans and um if you ever look up the International if you ever look up the International if you ever look up the International Phonetic alphabets you can see a catalog Phonetic alphabets you can see a catalog Phonetic alphabets you can see a catalog of these sounds and certain languages of these sounds and certain languages of these sounds and certain languages have a a phon invent have a a phon invent have a a phon invent um for the most part it's it's like all um for the most part it's it's like all um for the most part it's it's like all human things it's kind of a fuzzy thing human things it's kind of a fuzzy thing human things it's kind of a fuzzy thing and it changes over time or over regions and it changes over time or over regions and it changes over time or over regions and whatnot but so this acoustic model and whatnot but so this acoustic model and whatnot but so this acoustic model produces phon produces phon produces phon distributions and then the other part distributions and then the other part distributions and then the other part the language model which uh has like the language model which uh has like the language model which uh has like pronunciations for words and which words pronunciations for words and which words pronunciations for words and which words follow other words that's what you train follow other words that's what you train follow other words that's what you train so raspby allowed you to not only say so raspby allowed you to not only say so raspby allowed you to not only say What sentences you wanted to um be What sentences you wanted to um be What sentences you wanted to um be recognized but also you could provide recognized but also you could provide recognized but also you could provide custom pronunciation custom pronunciation custom pronunciation for words so that was one way to for words so that was one way to for words so that was one way to accommodate accents was to um alter the accommodate accents was to um alter the accommodate accents was to um alter the pronunciations of certain words to pronunciations of certain words to pronunciations of certain words to include phones that are not include phones that are not include phones that are not typical uh that's actually funny because typical uh that's actually funny because typical uh that's actually funny because not only is our house a non uh is a not only is our house a non uh is a not only is our house a non uh is a mixed accent house uh but my wife speaks mixed accent house uh but my wife speaks mixed accent house uh but my wife speaks uh Northern Ender belly or Zulu which uh Northern Ender belly or Zulu which uh Northern Ender belly or Zulu which includes click three different clicks oh includes click three different clicks oh includes click three different clicks oh interesting are those phon names interesting are those phon names interesting are those phon names included as well could she yell at The included as well could she yell at The included as well could she yell at The Voice Assistant with a click and would Voice Assistant with a click and would Voice Assistant with a click and would still understand her so those phone still understand her so those phone still understand her so those phone names are in the International Phonetic names are in the International Phonetic names are in the International Phonetic Alphabet but they are not typically in Alphabet but they are not typically in Alphabet but they are not typically in an English model that I've ever seen as an English model that I've ever seen as an English model that I've ever seen as possibilities yeah but so if I have a possibilities yeah but so if I have a possibilities yeah but so if I have a house full of different folks from house full of different folks from house full of different folks from different places all over the place I'd different places all over the place I'd different places all over the place I'd need a mixed model yeah you would need need a mixed model yeah you would need need a mixed model yeah you would need some kind of mixed model or or need some kind of mixed model or or need some kind of mixed model or or need multiple models running

multiple models running multiple models running simultaneously oh you can do that too simultaneously oh you can do that too simultaneously oh you can do that too you could could do that with a which one you could could do that with a which one you could could do that with a which one has the highest probability kind of has the highest probability kind of has the highest probability kind of situation that would be interesting situation that would be interesting situation that would be interesting because I was noticing that there's lots because I was noticing that there's lots because I was noticing that there's lots of different Mandarin models and stuff of different Mandarin models and stuff of different Mandarin models and stuff and lots of folks are inous that are and lots of folks are inous that are and lots of folks are inous that are could be speaking English and Chinese at could be speaking English and Chinese at could be speaking English and Chinese at the same time yeah we've tried to to the same time yeah we've tried to to the same time yeah we've tried to to make things work for multilingual make things work for multilingual make things work for multilingual households uh with with the The Voice households uh with with the The Voice households uh with with the The Voice Assistant stack it's It's Tricky um but Assistant stack it's It's Tricky um but Assistant stack it's It's Tricky um but one thing we're we're wanting to add one thing we're we're wanting to add one thing we're we're wanting to add soon is the ability to have um a soon is the ability to have um a soon is the ability to have um a different wake word would trigger a different wake word would trigger a different wake word would trigger a different language pipeline in our different language pipeline in our different language pipeline in our system so oh that'd be a smart way to do system so oh that'd be a smart way to do system so oh that'd be a smart way to do it yeah then you don't have to play it yeah then you don't have to play it yeah then you don't have to play games with you know trying to guess a games with you know trying to guess a games with you know trying to guess a probability of a sentence and whatnot probability of a sentence and whatnot probability of a sentence and whatnot yeah bypass the whole thing and just if yeah bypass the whole thing and just if yeah bypass the whole thing and just if they say computer in a different they say computer in a different they say computer in a different language then it's like all right Branch language then it's like all right Branch language then it's like all right Branch right at the opening I feel like that's right at the opening I feel like that's right at the opening I feel like that's that's kind of the underlying philosophy that's kind of the underlying philosophy that's kind of the underlying philosophy of raspy has been you know just ask like of raspy has been you know just ask like of raspy has been you know just ask like why don't you just tell me the voice why don't you just tell me the voice why don't you just tell me the voice commands you want to to recognize you commands you want to to recognize you commands you want to to recognize you know kind of thing uh instead of and and know kind of thing uh instead of and and know kind of thing uh instead of and and this is a difference between you know this is a difference between you know this is a difference between you know you adapting to technology versus you adapting to technology versus you adapting to technology versus technology adapting to you I kind of technology adapting to you I kind of technology adapting to you I kind of felt like it was a middle ground where felt like it was a middle ground where felt like it was a middle ground where this was you teaching the technology how this was you teaching the technology how this was you teaching the technology how to to to behave right behave right behave right okay so here's maybe a little bit harder okay so here's maybe a little bit harder okay so here's maybe a little bit harder question and I asked paulus this as well question and I asked paulus this as well question and I asked paulus this as well how do you get not just home assistant how do you get not just home assistant how do you get not just home assistant at raspy to break into more than just at raspy to break into more than just at raspy to break into more than just the Enthusiast techie world is it the the Enthusiast techie world is it the the Enthusiast techie world is it the outof boox experience like maybe someone outof boox experience like maybe someone outof boox experience like maybe someone who's listening is a home assistant fan who's listening is a home assistant fan who's listening is a home assistant fan they didn't know raspy existed how much

they didn't know raspy existed how much they didn't know raspy existed how much tinkering and if they have to open emac tinkering and if they have to open emac tinkering and if they have to open emac that's that's too hard uh you know that's that's too hard uh you know that's that's too hard uh you know before someone is successful setting up before someone is successful setting up before someone is successful setting up an offline voice assistant so raspy I an offline voice assistant so raspy I an offline voice assistant so raspy I would say is for the tinkerers if you're would say is for the tinkerers if you're would say is for the tinkerers if you're if you're not part of that crowd um then if you're not part of that crowd um then if you're not part of that crowd um then you you you probably want to ignore that you you you probably want to ignore that you you you probably want to ignore that for for the time being the out of- boox for for the time being the out of- boox for for the time being the out of- boox experience absolutely I think is the big experience absolutely I think is the big experience absolutely I think is the big differentiator and that's where on the differentiator and that's where on the differentiator and that's where on the home assistant side uh we've done a much home assistant side uh we've done a much home assistant side uh we've done a much better job with that so on the home better job with that so on the home better job with that so on the home assistant side you you know I I've taken assistant side you you know I I've taken assistant side you you know I I've taken pieces from rasby and basically put them pieces from rasby and basically put them pieces from rasby and basically put them in the in the ecosystem so that it fits in the in the ecosystem so that it fits in the in the ecosystem so that it fits very nicely and has a nice out-of-box very nicely and has a nice out-of-box very nicely and has a nice out-of-box experience for people people the easiest experience for people people the easiest experience for people people the easiest out-of-box experience of of course is to out-of-box experience of of course is to out-of-box experience of of course is to buy the hardware that we put out this buy the hardware that we put out this buy the hardware that we put out this home assistant voice we're calling it home assistant voice we're calling it home assistant voice we're calling it preview Edition because well the preview Edition because well the preview Edition because well the hardware is is very nice and and done hardware is is very nice and and done hardware is is very nice and and done these softwares for some people um is these softwares for some people um is these softwares for some people um is still a work in progress so we're kind still a work in progress so we're kind still a work in progress so we're kind of saying come on come on this journey of saying come on come on this journey of saying come on come on this journey with us yeah but the outof boox with us yeah but the outof boox with us yeah but the outof boox experience for this piece of Hardware um experience for this piece of Hardware um experience for this piece of Hardware um and for others that are coming from and for others that are coming from and for others that are coming from other companies is you you plug it in other companies is you you plug it in other companies is you you plug it in you you pair it with your home assistant you you pair it with your home assistant you you pair it with your home assistant and then you go and so that that I think and then you go and so that that I think and then you go and so that that I think is a a major leap uh from previously is a a major leap uh from previously is a a major leap uh from previously where you'd start with you know flash an where you'd start with you know flash an where you'd start with you know flash an SD card right for your your Raspberry Pi SD card right for your your Raspberry Pi SD card right for your your Raspberry Pi and and so and and so and and so on so is raspby the uh the underlying

on so is raspby the uh the underlying on so is raspby the uh the underlying product underneath home assistant product underneath home assistant product underneath home assistant assist I I would say ideas from it are assist I I would say ideas from it are assist I I would say ideas from it are but raspy itself is is the project that but raspy itself is is the project that but raspy itself is is the project that I I I had before coming to had before coming to had before coming to nabasa okay and so I I've taken all the nabasa okay and so I I've taken all the nabasa okay and so I I've taken all the lessons learned from that and put them lessons learned from that and put them lessons learned from that and put them forward into forward into forward into assist okay and is assist still open assist okay and is assist still open assist okay and is assist still open source or is assist not open source yeah source or is assist not open source yeah source or is assist not open source yeah yeah it's all open source under the yeah it's all open source under the yeah it's all open source under the raspy name uh it's it's kind of a raspy name uh it's it's kind of a raspy name uh it's it's kind of a branding thing a little bit but it's branding thing a little bit but it's branding thing a little bit but it's it's really under the assist I I would it's really under the assist I I would it's really under the assist I I would say just assist and and raspy can be say just assist and and raspy can be say just assist and and raspy can be seen as separate things with one being a seen as separate things with one being a seen as separate things with one being a predecessor is probably a better thing predecessor is probably a better thing predecessor is probably a better thing the reason I ask is I'm trying to the reason I ask is I'm trying to the reason I ask is I'm trying to understand how one can go and explore I understand how one can go and explore I understand how one can go and explore I mean certainly you can go to nabuk mean certainly you can go to nabuk mean certainly you can go to nabuk kasa.com kasa.com kasa.com nasa.com you can click on assist you can nasa.com you can click on assist you can nasa.com you can click on assist you can learn all about it and you can also learn all about it and you can also learn all about it and you can also surprisingly uh connect uh Alexa and surprisingly uh connect uh Alexa and surprisingly uh connect uh Alexa and Google Assistant to this if you want as Google Assistant to this if you want as Google Assistant to this if you want as well so you really are about choice I well so you really are about choice I well so you really are about choice I thought it was about avoiding those thought it was about avoiding those thought it was about avoiding those things but really you're letting anyone things but really you're letting anyone things but really you're letting anyone use anything they want however they want use anything they want however they want use anything they want however they want yeah so you can you can hook up your yeah so you can you can hook up your yeah so you can you can hook up your Alexa and uh speak to it there are some Alexa and uh speak to it there are some Alexa and uh speak to it there are some things you can't do just because Amazon things you can't do just because Amazon things you can't do just because Amazon won't let us so for example if you're won't let us so for example if you're won't let us so for example if you're running a cyst with with our Hardware or running a cyst with with our Hardware or running a cyst with with our Hardware or there's there's tons of of ESP based there's there's tons of of ESP based there's there's tons of of ESP based projects people can explore at Raspberry projects people can explore at Raspberry projects people can explore at Raspberry Pi based wantes to make a voice Pi based wantes to make a voice Pi based wantes to make a voice assistant um you can you can go in assistant um you can you can go in assistant um you can you can go in create an Automation in home assistant create an Automation in home assistant create an Automation in home assistant and say whenever I say this voice and say whenever I say this voice and say whenever I say this voice command do this and you can put whatever command do this and you can put whatever command do this and you can put whatever you want in there and when you say it you want in there and when you say it you want in there and when you say it it'll happen uh Alexa won't let us do

it'll happen uh Alexa won't let us do it'll happen uh Alexa won't let us do that because we don't get to see the that because we don't get to see the that because we don't get to see the text of the voice command it just does text of the voice command it just does text of the voice command it just does its thing it's all closed off right its thing it's all closed off right its thing it's all closed off right right right right okay so um I'm I'm in right right right okay so um I'm I'm in right right right okay so um I'm I'm in my house I've got home assistant I've my house I've got home assistant I've my house I've got home assistant I've got no internet because the internet's got no internet because the internet's got no internet because the internet's down I've got home assistant running on down I've got home assistant running on down I've got home assistant running on either a Raspberry Pi or in a Docker either a Raspberry Pi or in a Docker either a Raspberry Pi or in a Docker container inside my house we have no container inside my house we have no container inside my house we have no external talk where is the speaker what external talk where is the speaker what external talk where is the speaker what does it look like how many do I need to does it look like how many do I need to does it look like how many do I need to get and walk me through that process I'm get and walk me through that process I'm get and walk me through that process I'm in my house and I'm yelling to turn the in my house and I'm yelling to turn the in my house and I'm yelling to turn the lights on okay so you're going to have lights on okay so you're going to have lights on okay so you're going to have to have some kind of voice assistant to have some kind of voice assistant to have some kind of voice assistant Hardware uh like the The Voice PE or Hardware uh like the The Voice PE or Hardware uh like the The Voice PE or preview Edition I mentioned there are preview Edition I mentioned there are preview Edition I mentioned there are Raspberry Pi based Solutions or there Raspberry Pi based Solutions or there Raspberry Pi based Solutions or there are ESP home based ones so you you are ESP home based ones so you you are ESP home based ones so you you probably want one in each probably want one in each probably want one in each room and uh these these what they do is room and uh these these what they do is room and uh these these what they do is they'll they'll sit there waiting for they'll they'll sit there waiting for they'll they'll sit there waiting for the Wake word when you speak to it or the Wake word when you speak to it or the Wake word when you speak to it or when you say the Wake word it will when you say the Wake word it will when you say the Wake word it will stream the audio to your home assistant stream the audio to your home assistant stream the audio to your home assistant server where it's processed so depending server where it's processed so depending server where it's processed so depending on what you have as the speech detex on what you have as the speech detex on what you have as the speech detex processor there depends on what happens processor there depends on what happens processor there depends on what happens next okay so it's not listening at all next okay so it's not listening at all next okay so it's not listening at all times it is it isn't like we always talk times it is it isn't like we always talk times it is it isn't like we always talk about that idea like is it listening about that idea like is it listening about that idea like is it listening versus is it not right can you help me versus is it not right can you help me versus is it not right can you help me understand that because i' I've still understand that because i' I've still understand that because i' I've still struggled to explain to non-technical struggled to explain to non-technical struggled to explain to non-technical parent that listening to everything you parent that listening to everything you parent that listening to everything you say is different from a listening for a say is different from a listening for a say is different from a listening for a hard K in Alexa yeah it's a it's a hard K in Alexa yeah it's a it's a hard K in Alexa yeah it's a it's a different thing so uh basically the it different thing so uh basically the it different thing so uh basically the it will sit there keeping only just enough will sit there keeping only just enough will sit there keeping only just enough audio to check whether or not the Wake audio to check whether or not the Wake audio to check whether or not the Wake word is there so that's that's the only word is there so that's that's the only word is there so that's that's the only audio that's ever kept and that's

audio that's ever kept and that's audio that's ever kept and that's immediately thrown away and the moment immediately thrown away and the moment immediately thrown away and the moment it detects that that you sound you said it detects that that you sound you said it detects that that you sound you said the Wake word now it starts pushing that the Wake word now it starts pushing that the Wake word now it starts pushing that data out to the home assistant server in data out to the home assistant server in data out to the home assistant server in our case or in Alexa's case the internet our case or in Alexa's case the internet our case or in Alexa's case the internet yeah or in this case yeah inward yeah yeah or in this case yeah inward yeah yeah or in this case yeah inward yeah right if I have if I have the the voice right if I have if I have the the voice right if I have if I have the the voice assistant voice PE Puck sitting on my assistant voice PE Puck sitting on my assistant voice PE Puck sitting on my desk it's on my internal iot network desk it's on my internal iot network desk it's on my internal iot network with my home assistant yeah and it just with my home assistant yeah and it just with my home assistant yeah and it just bounces on the intet yep okay cool yeah bounces on the intet yep okay cool yeah bounces on the intet yep okay cool yeah so now the audio makes it to your home so now the audio makes it to your home so now the audio makes it to your home assistant server so now we got to do assistant server so now we got to do assistant server so now we got to do something with it and uh if you're something with it and uh if you're something with it and uh if you're running fully locally you have a couple running fully locally you have a couple running fully locally you have a couple options you you install the home options you you install the home options you you install the home assistant add-on for the handle speech assistant add-on for the handle speech assistant add-on for the handle speech to text there are a number of them out to text there are a number of them out to text there are a number of them out there we have one for whisper so you can there we have one for whisper so you can there we have one for whisper so you can run open AI whisper uh with with run open AI whisper uh with with run open AI whisper uh with with whatever model you choose uh that will whatever model you choose uh that will whatever model you choose uh that will run on your Hardware this is the tricky run on your Hardware this is the tricky run on your Hardware this is the tricky part so if you have if if you have a a part so if you have if if you have a a part so if you have if if you have a a beefy server or you have a a a graphics beefy server or you have a a a graphics beefy server or you have a a a graphics processor and you know what you're doing processor and you know what you're doing processor and you know what you're doing you can run these very large whisper you can run these very large whisper you can run these very large whisper models and you can get you know data models and you can get you know data models and you can get you know data center quality transcriptions out of it center quality transcriptions out of it center quality transcriptions out of it but if you're running something on a but if you're running something on a but if you're running something on a Raspberry Pi 4 then you're not going to Raspberry Pi 4 then you're not going to Raspberry Pi 4 then you're not going to be able to run the full strength models be able to run the full strength models be able to run the full strength models and um that's where I'm I'm creating an and um that's where I'm I'm creating an and um that's where I'm I'm creating an add-on right now actually that takes the add-on right now actually that takes the add-on right now actually that takes the tech from rasby puts this all into a tech from rasby puts this all into a tech from rasby puts this all into a nice little home assistant uh add-on so nice little home assistant uh add-on so nice little home assistant uh add-on so people can uh install it it

people can uh install it it people can uh install it it automatically trains itself on your uh automatically trains itself on your uh automatically trains itself on your uh your entity and and device names in your your entity and and device names in your your entity and and device names in your house with a limited set of voice house with a limited set of voice house with a limited set of voice commands all out of the box ready to go commands all out of the box ready to go commands all out of the box ready to go it it does mean you can't say anything it it does mean you can't say anything it it does mean you can't say anything to it although you can add to your own to it although you can add to your own to it although you can add to your own automations and such and it trains on automations and such and it trains on automations and such and it trains on those yeah so that's a real challenge those yeah so that's a real challenge those yeah so that's a real challenge because like right now I feel like uh because like right now I feel like uh because like right now I feel like uh you know I have mixed feelings about you know I have mixed feelings about you know I have mixed feelings about llms but I also are very often I'm llms but I also are very often I'm llms but I also are very often I'm watching a movie and I want to know you watching a movie and I want to know you watching a movie and I want to know you know how tall Brad Pit is so like Alexa know how tall Brad Pit is so like Alexa know how tall Brad Pit is so like Alexa for me is basically turning lights on for me is basically turning lights on for me is basically turning lights on and asking how tall an actor is ah so we and asking how tall an actor is ah so we and asking how tall an actor is ah so we have a we have a great solution for this have a we have a great solution for this have a we have a great solution for this actually um yeah what you can do is uh actually um yeah what you can do is uh actually um yeah what you can do is uh if I mean you need to be running if I mean you need to be running if I mean you need to be running something that can transcribe you know something that can transcribe you know something that can transcribe you know how tall is Brad Pit or or something how tall is Brad Pit or or something how tall is Brad Pit or or something like that but uh we have a fallback like that but uh we have a fallback like that but uh we have a fallback system so if if assist is unable to system so if if assist is unable to system so if if assist is unable to handle your command locally then it will handle your command locally then it will handle your command locally then it will fall back to another another thing which fall back to another another thing which fall back to another another thing which you can set as an llm or something more you can set as an llm or something more you can set as an llm or something more powerful so a lot of people are doing powerful so a lot of people are doing powerful so a lot of people are doing this with chat GPT or uh Claude or or this with chat GPT or uh Claude or or this with chat GPT or uh Claude or or Google gem and I so if you say if you Google gem and I so if you say if you Google gem and I so if you say if you say turn off the lights it stays say turn off the lights it stays say turn off the lights it stays completely local but if you ask a completely local but if you ask a completely local but if you ask a question outside of that scope then it question outside of that scope then it question outside of that scope then it falls back and goes out to the llm ah I falls back and goes out to the llm ah I falls back and goes out to the llm ah I wonder if it could fall back to a local wonder if it could fall back to a local wonder if it could fall back to a local slm it should be able to um we have Ama slm it should be able to um we have Ama slm it should be able to um we have Ama support in there so so you could run a support in there so so you could run a support in there so so you could run a little llama slm or something like that little llama slm or something like that little llama slm or something like that that's cool you know my I used to joke that's cool you know my I used to joke that's cool you know my I used to joke with my dad that because my dad's like a

with my dad that because my dad's like a with my dad that because my dad's like a woodworker and a not a computer guy so woodworker and a not a computer guy so woodworker and a not a computer guy so like if the if the zombies ever came I like if the if the zombies ever came I like if the if the zombies ever came I would not be very good like there's not would not be very good like there's not would not be very good like there's not a lot of software Engineers on The a lot of software Engineers on The a lot of software Engineers on The Walking Dead but I feel like the home Walking Dead but I feel like the home Walking Dead but I feel like the home assistant crowd can you imagine a show assistant crowd can you imagine a show assistant crowd can you imagine a show where it's like the zombies have where it's like the zombies have where it's like the zombies have destroyed the world there's no internet destroyed the world there's no internet destroyed the world there's no internet but all the home assistant users have a but all the home assistant users have a but all the home assistant users have a lovely voice model running they've lovely voice model running they've lovely voice model running they've downloaded complete copies of Wikipedia downloaded complete copies of Wikipedia downloaded complete copies of Wikipedia and they're still happily chatting away and they're still happily chatting away and they're still happily chatting away in an entirely local intranet uh with in an entirely local intranet uh with in an entirely local intranet uh with automation of everything including I automation of everything including I automation of everything including I assume you know zombie shooting Nerf assume you know zombie shooting Nerf assume you know zombie shooting Nerf guns outside something like that yeah guns outside something like that yeah guns outside something like that yeah I mean I love the idea of of making use I mean I love the idea of of making use I mean I love the idea of of making use of what is already in your house I mean of what is already in your house I mean of what is already in your house I mean You' you've got a dozen raspberry pies You' you've got a dozen raspberry pies You' you've got a dozen raspberry pies already right yeah you know being able already right yeah you know being able already right yeah you know being able to to make use of those and run to to make use of those and run to to make use of those and run something uh it's it seems like a better something uh it's it seems like a better something uh it's it seems like a better option than adding another subscription option than adding another subscription option than adding another subscription or another you know giant thing but or another you know giant thing but or another you know giant thing but there are some there are some people too there are some there are some people too there are some there are some people too where it's it's a bigger deal than just where it's it's a bigger deal than just where it's it's a bigger deal than just a you know preference of cloud versus a you know preference of cloud versus a you know preference of cloud versus not I've I've had conversations with not I've I've had conversations with not I've I've had conversations with people who are quadriplegic for example people who are quadriplegic for example people who are quadriplegic for example and uh they had to like they had to add and uh they had to like they had to add and uh they had to like they had to add um Alexa and Google home and Siri all of um Alexa and Google home and Siri all of um Alexa and Google home and Siri all of these things because if any one of those these things because if any one of those these things because if any one of those Services went out they were you know Services went out they were you know Services went out they were you know they they were in a a bad situation they they were in a a bad situation they they were in a a bad situation they'd have to call somebody or try to they'd have to call somebody or try to they'd have to call somebody or try to get a hold of somebody and so they were get a hold of somebody and so they were get a hold of somebody and so they were very interested in raspy and and local very interested in raspy and and local very interested in raspy and and local voice because they needed something more voice because they needed something more voice because they needed something more robust just just for their day-to-day

robust just just for their day-to-day robust just just for their day-to-day activities activities activities um so I think you know to me that's um so I think you know to me that's um so I think you know to me that's that's a really cool cool intersection that's a really cool cool intersection that's a really cool cool intersection where you you have techies who are where you you have techies who are where you you have techies who are interested for privacy reasons and you interested for privacy reasons and you interested for privacy reasons and you have people who need it for just to live have people who need it for just to live have people who need it for just to live their their their lives so I'm over here at lives so I'm over here at lives so I'm over here at home-assistant.io home-assistant.io home-assistant.io v-p I'll put a link in the show notes v-p I'll put a link in the show notes v-p I'll put a link in the show notes and it says I can buy the home assistant and it says I can buy the home assistant and it says I can buy the home assistant voice preview Edition now and it's just voice preview Edition now and it's just voice preview Edition now and it's just 59 bucks yep that's amazing and it's not 59 bucks yep that's amazing and it's not 59 bucks yep that's amazing and it's not it's include I mean it's included in my it's include I mean it's included in my it's include I mean it's included in my existing subscription I can keep running existing subscription I can keep running existing subscription I can keep running the home assistant or I can just pay the home assistant or I can just pay the home assistant or I can just pay nabasa for cloud support but uh there's nabasa for cloud support but uh there's nabasa for cloud support but uh there's no hidden fee underneath this no so you no hidden fee underneath this no so you no hidden fee underneath this no so you can if you have a home home assistant can if you have a home home assistant can if you have a home home assistant subscription uh home assist and Cloud subscription uh home assist and Cloud subscription uh home assist and Cloud subscription you can hook it up and you subscription you can hook it up and you subscription you can hook it up and you can do voice voice to text and text to can do voice voice to text and text to can do voice voice to text and text to voice through the nabasa cloud which voice through the nabasa cloud which voice through the nabasa cloud which really goes out to Microsoft really goes out to Microsoft really goes out to Microsoft Azure uh and that works for a lot of Azure uh and that works for a lot of Azure uh and that works for a lot of people there's huge language support if people there's huge language support if people there's huge language support if you prefer not to use a cloud then we you prefer not to use a cloud then we you prefer not to use a cloud then we have tutorials for that um we even have have tutorials for that um we even have have tutorials for that um we even have a little wizard that will guide you a little wizard that will guide you a little wizard that will guide you through that yeah and there's a whole through that yeah and there's a whole through that yeah and there's a whole section on contributing to the voice section on contributing to the voice section on contributing to the voice initiative so if you're a native speaker initiative so if you're a native speaker initiative so if you're a native speaker of a voice that is not supported I went of a voice that is not supported I went of a voice that is not supported I went and I selected Zulu it's not currently and I selected Zulu it's not currently and I selected Zulu it's not currently supported you can go and create and supported you can go and create and supported you can go and create and donate a piper voice yes so Piper is the donate a piper voice yes so Piper is the donate a piper voice yes so Piper is the is a text to speech system that I I want is a text to speech system that I I want is a text to speech system that I I want to say I built I I built Piper but the to say I built I I built Piper but the to say I built I I built Piper but the model underlying model was built by model underlying model was built by model underlying model was built by somebody else um uh it's called a vits

somebody else um uh it's called a vits somebody else um uh it's called a vits or vits model and so I I didn't research or vits model and so I I didn't research or vits model and so I I didn't research and design that model or anything but and design that model or anything but and design that model or anything but yeah I trained a bunch of these Piper yeah I trained a bunch of these Piper yeah I trained a bunch of these Piper voices and I still get uh I get voices and I still get uh I get voices and I still get uh I get volunteers I have probably a dozen volunteers I have probably a dozen volunteers I have probably a dozen volunteers now that are reading volunteers now that are reading volunteers now that are reading sentences online um we've we've sentences online um we've we've sentences online um we've we've collected about two to three hours worth collected about two to three hours worth collected about two to three hours worth of recordings from each of these people of recordings from each of these people of recordings from each of these people and that's enough to train a voice and that's enough to train a voice and that's enough to train a voice that's that's good enough uh to give that's that's good enough uh to give that's that's good enough uh to give responses to voice commands and I tried responses to voice commands and I tried responses to voice commands and I tried I'm trying to get as many languages as I I'm trying to get as many languages as I I'm trying to get as many languages as I possibly can nabasa will pay for uh possibly can nabasa will pay for uh possibly can nabasa will pay for uh recording studio time for certain recording studio time for certain recording studio time for certain languages and then we take that data and languages and then we take that data and languages and then we take that data and put it out in the public domain so you put it out in the public domain so you put it out in the public domain so you know a future future Texas speech know a future future Texas speech know a future future Texas speech projects will hopefully pick that up and projects will hopefully pick that up and projects will hopefully pick that up and say oh there you know maybe in the say oh there you know maybe in the say oh there you know maybe in the future there's a there's a Zulu data set future there's a there's a Zulu data set future there's a there's a Zulu data set right there I'll just I'll train a a right there I'll just I'll train a a right there I'll just I'll train a a Zulu voice you know why not yeah yeah Zulu voice you know why not yeah yeah Zulu voice you know why not yeah yeah and there's also public data sets like and there's also public data sets like and there's also public data sets like LibriVox and Mozilla common voice those LibriVox and Mozilla common voice those LibriVox and Mozilla common voice those um so common voice has a different um so common voice has a different um so common voice has a different mission in text of speech uh this is mission in text of speech uh this is mission in text of speech uh this is where the computer speaks back to you where the computer speaks back to you where the computer speaks back to you right uh you you want to have usually right uh you you want to have usually right uh you you want to have usually few speakers and you want to have a lot few speakers and you want to have a lot few speakers and you want to have a lot of data from one speaker in a quiet nice of data from one speaker in a quiet nice of data from one speaker in a quiet nice quiet environment um speech detects you quiet environment um speech detects you quiet environment um speech detects you kind of want the opposite you want a ton kind of want the opposite you want a ton kind of want the opposite you want a ton of people in very noisy environments of people in very noisy environments of people in very noisy environments because that's going to be real world because that's going to be real world because that's going to be real world real world conditions toh to hear voice real world conditions toh to hear voice real world conditions toh to hear voice commands and while LibriVox sounds like commands and while LibriVox sounds like commands and while LibriVox sounds like it would be perfect this is people it would be perfect this is people it would be perfect this is people reading audiobooks in the public domain reading audiobooks in the public domain reading audiobooks in the public domain I have actually found it to be I have actually found it to be I have actually found it to be Troublesome with training text to speech Troublesome with training text to speech Troublesome with training text to speech models because uh narrative text almost

models because uh narrative text almost models because uh narrative text almost never contains short phrases very short never contains short phrases very short never contains short phrases very short and so the neural network models uh they and so the neural network models uh they and so the neural network models uh they end up basically falling to pieces if end up basically falling to pieces if end up basically falling to pieces if you respond yes no or you know turned on you respond yes no or you know turned on you respond yes no or you know turned on the light they they need a full the light they they need a full the light they they need a full practically a full paragraph before they practically a full paragraph before they practically a full paragraph before they start sounding start sounding start sounding correct uh so this is part of my Piper correct uh so this is part of my Piper correct uh so this is part of my Piper training sets as I have people read a training sets as I have people read a training sets as I have people read a bunch of short phrases a bunch of single bunch of short phrases a bunch of single bunch of short phrases a bunch of single numbers and things like that well it numbers and things like that well it numbers and things like that well it sounds like it's a really interesting sounds like it's a really interesting sounds like it's a really interesting time to be in this in the voice space time to be in this in the voice space time to be in this in the voice space especially in the open source voice especially in the open source voice especially in the open source voice space especially given all the talk space especially given all the talk space especially given all the talk about llms that are happening right now about llms that are happening right now about llms that are happening right now yeah uh I I feel like we're kind of in a yeah uh I I feel like we're kind of in a yeah uh I I feel like we're kind of in a very unique situation so very unique situation so very unique situation so we on the one hand The Voice existing we on the one hand The Voice existing we on the one hand The Voice existing voice assistants are kind of crumbling voice assistants are kind of crumbling voice assistants are kind of crumbling like Alexa and Google and stuff they're like Alexa and Google and stuff they're like Alexa and Google and stuff they're getting worse uh as as companies aren't getting worse uh as as companies aren't getting worse uh as as companies aren't making any money off of it I guess and making any money off of it I guess and making any money off of it I guess and at the same time everyone's getting at the same time everyone's getting at the same time everyone's getting excited exed about llms and uh we excited exed about llms and uh we excited exed about llms and uh we provide a path where you can just hook provide a path where you can just hook provide a path where you can just hook hook it right up to chat GPT if you want hook it right up to chat GPT if you want hook it right up to chat GPT if you want so right now you could if you had the so right now you could if you had the so right now you could if you had the home assistant Cloud subscription and a home assistant Cloud subscription and a home assistant Cloud subscription and a open AI subscription you can have a open AI subscription you can have a open AI subscription you can have a fully you know a full Voice Assistant fully you know a full Voice Assistant fully you know a full Voice Assistant that goes uh straight up to chat GPT and that goes uh straight up to chat GPT and that goes uh straight up to chat GPT and controls your house yeah it does seem to controls your house yeah it does seem to controls your house yeah it does seem to me though that uh that's wasteful from a me though that uh that's wasteful from a me though that uh that's wasteful from a power perspective and that's a power perspective and that's a power perspective and that's a combination of like uh home assistant combination of like uh home assistant combination of like uh home assistant yellow or home assistant connect yellow or home assistant connect yellow or home assistant connect with the voice preview Edition would be with the voice preview Edition would be with the voice preview Edition would be the most energy efficient and most the most energy efficient and most the most energy efficient and most private and clean way to get involved in

private and clean way to get involved in private and clean way to get involved in this space yeah um if you're really this space yeah um if you're really this space yeah um if you're really wanting to do the the llm stuff though wanting to do the the llm stuff though wanting to do the the llm stuff though locally um and you're wanting to be locally um and you're wanting to be locally um and you're wanting to be performant and power efficient I there performant and power efficient I there performant and power efficient I there are a number of choices but they're are a number of choices but they're are a number of choices but they're they're not they're pretty expensive they're not they're pretty expensive they're not they're pretty expensive like um Nvidia has their Jets and boxes like um Nvidia has their Jets and boxes like um Nvidia has their Jets and boxes that you can buy I mean but the ones the that you can buy I mean but the ones the that you can buy I mean but the ones the ones that you really want are you know ones that you really want are you know ones that you really want are you know two to5 thousand kind of deal um and two to5 thousand kind of deal um and two to5 thousand kind of deal um and they're not very plug andplay with with they're not very plug andplay with with they're not very plug andplay with with home assistant I feel like that's kind home assistant I feel like that's kind home assistant I feel like that's kind of a missing piece I would love to see of a missing piece I would love to see of a missing piece I would love to see in sometime in the future where I mean in sometime in the future where I mean in sometime in the future where I mean people could have an AI box sitting in people could have an AI box sitting in people could have an AI box sitting in their house right that that right runs their house right that that right runs their house right that that right runs these kinds of things um and it' be these kinds of things um and it' be these kinds of things um and it' be great if if yeah if we could hook it up great if if yeah if we could hook it up great if if yeah if we could hook it up to Home assistant yeah I've got a to Home assistant yeah I've got a to Home assistant yeah I've got a co-pilot PC a Microsoft co-pilot PC co-pilot PC a Microsoft co-pilot PC co-pilot PC a Microsoft co-pilot PC which is a Snapdragon and it has a 40 which is a Snapdragon and it has a 40 which is a Snapdragon and it has a 40 Topps npu and I can run L 32 on the npu Topps npu and I can run L 32 on the npu Topps npu and I can run L 32 on the npu in like 20 watts so like low powered and in like 20 watts so like low powered and in like 20 watts so like low powered and then using something like LM Studio then using something like LM Studio then using something like LM Studio orama you can then expose it out to the orama you can then expose it out to the orama you can then expose it out to the local network as an open AI compatible local network as an open AI compatible local network as an open AI compatible endpoint and it's very Cutting Edge but endpoint and it's very Cutting Edge but endpoint and it's very Cutting Edge but I feel like it doesn't need to be it I feel like it doesn't need to be it I feel like it doesn't need to be it could be very normal for someone to have could be very normal for someone to have could be very normal for someone to have a small npu powered box in their house a small npu powered box in their house a small npu powered box in their house in the future yeah and I think doing in the future yeah and I think doing in the future yeah and I think doing that kind of thing also changes a lot of that kind of thing also changes a lot of that kind of thing also changes a lot of how you would um have home assistant or how you would um have home assistant or how you would um have home assistant or or other software use the AI so uh one or other software use the AI so uh one or other software use the AI so uh one example is if you had a VM model of with example is if you had a VM model of with example is if you had a VM model of with a vision language model you could have a

a vision language model you could have a a vision language model you could have a camera from your home assistant you know camera from your home assistant you know camera from your home assistant you know every five seconds hand it a snapshot every five seconds hand it a snapshot every five seconds hand it a snapshot and say what do you see what do you see and say what do you see what do you see and say what do you see what do you see what do you see right and this is not what do you see right and this is not what do you see right and this is not something you'd ever want to do with an something you'd ever want to do with an something you'd ever want to do with an open AI subscription because you'd burn open AI subscription because you'd burn open AI subscription because you'd burn through all your tokens you know in in a through all your tokens you know in in a through all your tokens you know in in a day or you'd rack up a huge bill but if day or you'd rack up a huge bill but if day or you'd rack up a huge bill but if you can if you can continuously ping a you can if you can continuously ping a you can if you can continuously ping a box that's sitting there in the house uh box that's sitting there in the house uh box that's sitting there in the house uh you could do lots of really cool things you could do lots of really cool things you could do lots of really cool things uh local uh local uh local very cool well thank you so much for very cool well thank you so much for very cool well thank you so much for chatting with me Dr Michael Hansen from chatting with me Dr Michael Hansen from chatting with me Dr Michael Hansen from nabuk Casa there's a lot of really cool nabuk Casa there's a lot of really cool nabuk Casa there's a lot of really cool stuff happening in the space and you can stuff happening in the space and you can stuff happening in the space and you can go and explore it all at nabuk kasa.com go and explore it all at nabuk kasa.com go and explore it all at nabuk kasa.com or at the home assistant site and learn or at the home assistant site and learn or at the home assistant site and learn about raspy learn about assist and maybe about raspy learn about assist and maybe about raspy learn about assist and maybe buy some of this Hardware that they're buy some of this Hardware that they're buy some of this Hardware that they're putting together thank you so much yeah putting together thank you so much yeah putting together thank you so much yeah thank you for having me this has been thank you for having me this has been thank you for having me this has been another episode of Hansel minutes and another episode of Hansel minutes and another episode of Hansel minutes and we'll see you again next week