 Well, did you migrate the old data over just not for test I can Well, yeah, you should probably do that because I'm gonna need some data to test with I'm looking for like, you know Hey, my craft weight words. I have zero for male or female. So I have You know, yes and no But anyway, so yeah, we can get the real day even even if you come before the data would gonna need to tag all that right What's that? Well, you can actually even if you even if you show truckloads of daughter in there We're gonna need to tag those samples before though 10,000 that have already been tagged micro a mycroft or not Yeah, that's the only one but I can get those. Yeah, we have anything you say you wanted like male or female like masculine or feminine Yeah, well, I'm not gonna get that I guess not right away So the So once you get through the hardware headaches that we're involved in now and kind of get this eight plus two things back to Do normal how far away are we from having the wake word process automated? So, you know a new wakeward goes into the queue it gets processed it goes into the data set And then we have a nightly or a weekly or whatever the bill time is Builds that rebuilds the wakeward spotter test it against the current wakeward spotter to see if it's better or worse And if it's better to poison Right the answer to that question is as soon as we can get some data and the test URLs so I can test and Validate and then turn it on in production. So, you know Okay, and does that if that includes like the automated build and test and deploy process That includes the automated builds, but it doesn't automatically deploy them Okay, dude, does it automatically test it like is there a test criteria where we compare the new model? It didn't see existing model. That is correct. That's part of the process Yeah, it's really awesome. I'm looking forward to seeing it Yeah, and right now the threshold is like a thousand more samples of the smallest of the six buckets a Bucket being we have six buckets right now. They're divided by Well, huh, right now we have three buckets, which is is not in like So it is the wake word. It's not the wake where it's like the wake word I'm hoping to start getting things tagged male and female at some point and There's no reason why that couldn't be turned on in production like yesterday But yeah, so right now it'll be you know, it'll be a threshold of a certain number of the smallest of those three buckets being reached Okay, and so that'll give us a call to action and Well, that will kick off it'll kick off. What's called a hyper parameter training session Which will basically run a bunch of different hyper parameter values. It'll train a bunch of different models It'll test all those models against the last model that were the current production model and see if it's improved And if it's improved Then where it was before I when I dropped it was I needed to notify somebody via email And I'm pretty sure you had get send me some information about our email processing system So that's where I was at but right now I still have to modify it because it's making its decisions based upon pulling the data from a database And I have to modify it to pull that data from Chris's URLs And that's where I'm at. Hey, that's all that's all good technical info But I'm I'm excited about the ability to issue a call to action to our community come help us taxi Masking someone and so forth. I'm super excited about that. Thank you. Ken. Yep Yeah, you know this week and maybe if there's a tweak I need to make for Ken to the API I can do that in a different release, but you know, just but everything else is we want to start tagging data I can spend some time. We should probably start tagging data I'll work around the the URL API as you gave me for now, but obviously it's not going to scale that way But yeah for now Let's start getting them tagged at least on another axis, which is gender Because that's kind of where we're at in our process. Okay, so what I'm hearing is that we can turn that on today Should be able to Well, I need to do yeah, I need to do the deployment to production Which it'll take me several hours to do but yes, I can do that today so whenever that's done, could you send me a note so that we can We can test it internally first a little bit and then once we're happy with it Let's issue a call to action for our community to reach in and help us to get some of this stuff squared away I'm really excited to it's been a while, but we now have some places we can point people at who are non developers If they can come help so I'm excited about that. Okay. I have a minor nuance regarding the URL as I've seen so far the API I don't see how to Determine or how to relate a sample to a user and While it's not a concern immediately Ultimately, we may want to limit the number of samples per user So I don't know how to do that Yeah, what I need today is data in the test URLs and you guys can go and start tagging the other 115,000 samples on the gender axis and then any new samples that come in on is is not or like so Should we jump in the test and just tag tag a bunch of data quickly? Is that is that helpful or? We need to just get it into production. Have we validated a bit of work in test? I Need more data. I always like to validate. It's working in test But I'm flexible. I mean, we're you know, yeah My question I guess my question is is the problem that you need more data Or we don't actually know whether it works because we don't have any data because like I've tagged a bunch of samples already So I assume that there's something in there Yeah, so the answer to your question is the latter. There is no real data I mean, I think I pulled back the most on any of these was like 10 or 15 so I'm assuming they're hard-coded samples or something They're not hard-coded. They're they're actual, you know Me and Gazz and some other people doing some tagging and and running the some scripts against the tags to make designations So we have like 15 wake words that have been tagged the same I craft Sounds about right. Okay by the internal team over the last, you know, couple of weeks Like but if those tags are good like clearly you're not going to train a model on it So like is that what we're trying to validate? What do we need to validate? So we are and then my question would become if we already have those and they're tagged But they're only tagged on a mycroft or not a mycroft. Why aren't they being tagged for gender as well? That's the next step and they would be if I continued to tag Well, whatever it takes to get that process flowing and getting that pipe filled I would recommend doing Yeah, so it sounds like Chris what you're saying is that the thing that's keeping us from moving on to the second level of tagging is Just having enough data for the algorithm to select those as the next things to be tagged well You not really so so yes, there's a Couple things one is we can we can pull over Whatever's in the precise database right now is which as far as which are a mycroft and which aren't The trouble becomes when we start the not hey mycrofts if we give those If they're noise for example Remember we entered we we added a noise layer That says it's a person speaking or not And the way that the UI works right now, that's the first layer The second layer is whether or not it's a mycroft and the third layer is anything after that basically So I can copy over a bunch of just data that's imprecise But there may be some noise in there and people we may be asking people if noise is a female or male So that could be potentially be a problem All right, it sounds like we need to have a discussion offline about this. Let's follow up on this right after this call It's getting a little bit too involved right now. Yeah But I I literally just tagged something as a masculine sample, so there should be at the very least one sample that has agenda now At least So if that doesn't work then clearly he's only burning Well, just because you tagged it doesn't mean it's been fine. It's been designated Which is the next step after you get to have multiple tags and it's with the same that agree in order to Yeah, so you're halfway there All right, okay Okay, let's go let's go in here because I'm I'm a little confused about exactly what the uh Because I know what we specced. I just don't know what state it's in right now I know that we weren't going to get to the full spec, you know before we started releasing things Um, but I need to you know, I want to discuss What exactly is the sequence, you know as things go through the tiger? But it sounds like we're pretty pretty close, but um, let's go through that after meeting and we'll figure that out So chris, what have you been up to? So yeah, so today, um, I started looking at the accounts and memberships and how that's going to change um, I submitted a pr with some changes to the database uh File structure, um, just a refactor and to prepare for adding a permission schema Which is my next thing to do is to add the schema. Um, I've designed Um, and then once the scheme is there, we can start to figure out where we want to add permissions What we want to add permissions to Um, so that's what I was doing. Um, we talked a little bit before this about maybe doing some more um, technical documentation about the architecture of this so I may take a step back Um and do that tomorrow Um, but then there's also the wake word thing depending on what we decide on there You know, I could spend some time deploying that tomorrow as well So, okay. Yeah, that seems likely. Um, all right. Uh, yes um Yeah, so another time, uh, if I um, the panicle stuff again, um, that's going Ken's going to patch part and and um, uh Putting Some of that into the devil's well putting it all into the devil's recipe as well So it was going up to the future if we need it. Um, we'll also just sort of making that a bit clearer and what, um What panicle I need to do, um Uh Did some debugging on the so the wi-fi Set up is all in the in the stable image now. Um, so if you if you get the image you could put it on your device and and start it up and set up wi-fi and and pair the device but You know, unless you've got working audio, then it's not going to be a very great experience, please Uh, but there we have There is an issue where it seems like core isn't starting until wi-fi connect finishes its process. So, um, It's been digging into that. Um With the help of both panicle and blue systems. Um Um And On the blue system side we caught up with them last night um, there's a The deed has done a camera skill already for us. So Um anyone has the camera module already? Um, I'll post that in the team chat uh, they're also Um Uh, I've been chatting with javas. Uh, because we've been doing these workshops for the catalan community. Um, and they've been looking at Uh, realize that we need to translate essentially translate the qml ui elements. Um, so most of the time most stuff is going to come from the skill itself But you know things like buttons and things make sense to actually You know have that coded in the in the ui rather than passing it as session data from the skill. Um So anyway looking at what we need to do To extend translation, you know localization stuff into the ui layer of the skill Okay, and there's a few different ways that we can go. Um, yeah, I think that's about it Okay, great Chris, uh, vera can we jump back to you for a minute and uh, feel us in on what dereca Had to say sure uh, so In matter most he says we put together and shipped a set of the latest laser cut housing and audio chamber So we can send it to ken um And 3d printed and shipped Speaker mounting parts to me when he gave me the parts the other day There were a couple parts missing so he shipped those to me um Working on a spec for sourcing the jst speaker wires, whatever those are um soldering a few speakers and crepping parts for more 3d printed prototypes and organizing the parts that are coming in for the uh, the build of the 300 the first 300 devices so as for myself, um I know I was out for most of today, but I have been working on sourcing the fans um is Turns out they're more expensive than we want them to be so i'm trying to find a supplier that can supply them less expensively and I had a good chat with kevin this morning about getting the next rev out Hopefully we'll have some good news about that this evening Um And uh, that's pretty much the update for me today Uh, josh, do you have any anything you want to add? No, the patent stuff Uh yesterday, I think went fairly well We'll kind of see what plays out there, um The uh And then yeah, I had to travel and they won't let me into the hospital So I am in a public park for a couple hours waiting for tests to be done I'll be back in cona tomorrow and I'll be back in the office first thing tomorrow morning local Okay, great, uh anything else we should bring up Not related to Okay, great. Well, then we'll call it there. Thanks everybody Um, we're gonna have a couple of follow-up calls immediately afterwards Stick around for that