Local Smart Home Voice Control Is AMAZING

Exactly two years ago, almost day by day, Home Assistant announced that they were going to start what they called the Year of the Voice, where throughout the year 2023 there was going to be a big change and focus on the addition of a local private vocal assistant, which did not really exist before. And throughout 2023 and all this year in 2024, the team added the basic elements of the voice at each release. But we have never had official material until today with the launch of the brand new edition of Home Assistant's Vocal Overview. The first official attempt to add a vocal assistant dedicated to Home Assistant. Let's see how it works. A total transparency as always, it is important to mention it. We are distributors and sellers of the P.E. voice via our store, which is of course a conflict of interest for this magazine. And so of course I want to bring my own point of view and my own turn and I think I can give my own unique information about the voice and nothing will ever stop me from saying what I want to say. But I encourage you to look at the reviews and consult other information from other creators, of which I am sure that your reception box is full today, so that you can get a complete idea before deciding if it is for you and your smart home. So I just wanted to mention that before moving on to the review. If you want to buy one, of course, I will have the links in the description. As mentioned, Home Assistant Voice P.E. is the first material from Nabucasa to tackle the addition of a Home Assistant voice assistant, which was one of the greatest features that was missing in a way compared to people coming from Amazon or Google Home. And that's all. It is a relatively small and compact unit in a more square and square design, finished in white with the translucent transparent plastic that we know of green and yellow. On the top, we have a clickable button surrounded by a rotary wheel for volume control, as well as a LED notification ring. And finally, a double set of microphones. At the back, we have a castrated hole housing the USB-C connector for power and a 3.5 mm audio jack for audio output. And on the side, a physical deaf switch that cuts the power of the microphones for people concerned about their private lives. And finally, at the front, there is a speaker grid housing the internal speaker. And on the bottom, a tab that you can break to access the port Inside, we have a Spanish 32-bit S3 processor accompanied by an XMOS processor, which manages the audio processing side and provides features such as echo cancellation, noise cancellation, as well as automatic gain. And it is this XO processor that really distinguishes it from the previous vocal equipment that we have seen in the past. The voice is of course executed by ESP Home, as one might expect, given that it comes from Nabucasa, which is incredible, because it means that many of the huge improvements they have made to the PE will be added to ESP Home so that everyone can use them, which is really cool, and this somehow reinforces their open source mission, as well as other pleasant touches, such as the feet of the case which are not glued and which can be completely removed, as well as a super easy disassembly. Finally, the price of detail of the Voice PE is 59 pounds, Once you have connected the Voice to the power supply and you have configured it, it will start blinking, which means it is ready to start the configuration process. And then, it's a very simple process to add it to Home Assistant directly from the Home Assistant application. Enter your Wi-Fi information, then the guide will guide you by pronouncing the alarm clock to activate it and that's about it. You will be asked how you want to use the voice in Home Assistant, either via Home Assistant Cloud if you are a subscriber to Nabucasa, or you can do it directly on your Home Assistant server. In Home Assistant, you will get a set of entities, including a multimedia player, a volume cursor, button events, commands for the detection of end of speech that allow you to modify the speed with which the listening stops. You also have commands for the OLED ring, the mute switch, the alarm clock and a selector for the alarm clock, among which you can choose from three options from the output. For those of you who do not know, the operation of Voice P.E. is as follows. There are in a way two elements in the vocal part, where you have a wake-up call or the sentence that is used to activate your Voice P.E., then you have the command or request, which is what you really want it to do or act on. And on Voice P.E., the wake-up call is made locally on the device, while the request will be managed by your Home Assistant server depending on how you want to configure it. I do not want to go into the technical details of a review like this, but I think it is important to talk about it a little, because many of you will naturally compare this to the S3 box or to the DIY voice attempt that we have been able to do in the past. Having a wake-up call on the device is a huge improvement over the S3 box, and you may be wondering how this can be possible when using the same wake-up call. And it seems that the audio processor really helps to obtain a vocal input as clean as possible so that the wake-up call model can do its best. It therefore performs this automatic gain as well as the cancellation of the echo and the suppression of the noise to clean the vocal input so that the wake-up call can do its best job it can. And it's a bit like if you put waste, you're going to get waste out of it. So if you pronounce the sentence and there is echo and noise in it and just a lot of echo in the room, then the model will have trouble and not work sometimes. But with the audio processor that is in the vocal PE, it means that you get an excellent input and that you will get excellent results. It's not perfect. There have certainly been strange occasions where he didn't recognize me by saying the wake-up word, which could be due to my accent maybe, and even Google Home is not perfect for my voice either. They sometimes fail, but I found that the wake-up call and Jarvis seemed to work better for my voice in particular, and I don't think I've had a single failure using this wake-up call. So maybe it's worth trying if you're on the same boat as me. As for the treatment of requests, I actually tried several different material options to perform the local vocal treatment with Home Assistant Cloud. And I was constantly patching from one to the other to get an idea of ​​how they work. By testing first on the Home Assistant Yellow or a Pi 4, I got about 2 to 3 seconds of vocal treatment on basic commands like turn on or off the lights and about 5 seconds for larger or longer requests. OK, Nabu, turn off the kitchen light. Turn off the lights. OK, now. I would say that all the integrated Home Assistant sentences tend to work well. I encountered occasional failures where it detected the word I was pronouncing incorrectly, but I noticed that there were more words missing than usual when it came to something more complex, like when you talked to the GPT cat, for example, which we will talk about in a second. But I guess the best way I can describe the local treatment on a Pi 4 or a Yellow is that I find myself talking to it more like you talk to your grandmother when she doesn't hear you, if you see what I mean. Like slightly slower and more pronounced to make sure they understand it, which is once again good for commands walk, stop. It works pretty well, but not for more complex requests. It can be quite easily resolved on the Yellow or a Pi 4 or a similar material if you are a Nabu Kaza subscriber who gives you access to the Home Assistant Cloud vocal treatment. Now, I know that some of you do not agree with the use of the Cloud for vocal commands and I totally respect that. But for those of you who don't mind, you get a really nice experience using Home Assistant Cloud, where they can run much more complex models much faster and they will produce much better results. And the vocal treatment side on Home Assistant Cloud is incredibly fast. We're talking about a few milliseconds to deal with the vocal side, even for more complex requests. Simple commands walk, stop. We're talking about something like 0.1 seconds to treat and do everything you need, which is absolutely crazy. And it seems very, very responsive. OK. Navi, turn off the kitchen lights. Turn off the lights. But if you don't want to run the Cloud and you want everything to be done locally and you have a slightly more powerful material than a Pi 4, you can get excellent results by doing everything locally. I use this mini computer, the iCoolCore R2 with a N95 processor, which is very low in power and not so new. There are now more powerful and more efficient chips. And from Whisper to the basic model, I was able to get 1.1 to 1.2 seconds to run everything for simple commands, which seems a lot. But when you actually use it, it seems very fast, very fast and much faster than anything Google or Amazon would do. OK. Nabu, turn off the kitchen lights. Turn off the lights. OK. Nabu, what time is it? 2.56 p.m. I think, conversely, the biggest compliment I can give is that with the basic model and above or with Home Assistant Cloud, I spoke completely naturally at my normal speed and the failure rate was very, very low. So, a small improvement of the material made a huge difference in speed and also in the precision of the voice that does it locally. And again, it is this audio processor that will work and that simply makes the vocal input as clean as possible so that the vocal model can do its job correctly. I was really blown away by the difference in precision here compared to the S3 box that I used before, for example. What is cool is that even if the local model makes a mistake in a word you said, which does not often happen, but if it does, if you have changed PT or another LLM configuration with Home Assistant, it saves you in a way. Thus, the LLM help option has just been added to this month's Home Assistant version. And during my voice tests, I realized how important this feature is for vocal PE. I think this is one of the most important features of the vocal PE edition. And it's really cool because there are times when I said something like turn on the ceiling and the local voice thought I had said turn on the ceiling height, for example. And the local treatment was like, well, I have no idea what the height is. And so he sent the request to ChaGPT and ChaGPT simply turned on the light I was talking about because it seemed to recognize that it was a mistake and since it only receives text, it probably thought it was a typo or something like that and so it finished the action totally correctly. And I thought it was really cool, like a kind of backup of your vocal model and so my experience using the local PE with local treatment on this little computer with ChaGPT was excellent. And I would definitely recommend it. It was a very good experience, at least for me. And if you want to know how much money I spent on the ChaGPT, I spent about $ 6 yesterday, a day where I asked a lot more questions than an average day. So, let's be generous and call it about $ 2 per month on the ChaGPT with the voice. And you can run a local LLM with Olama, which I also tried. It works pretty well too. However, you need more powerful equipment to run on your network The other huge advantage of the LLM function is that it transforms the vocal PE of a device that can only be used to turn on or off the devices and lights of your smart home and brings it so close to a Google or Amazon speaker because all of a sudden, it is able to answer all these random questions and requests that you may have, just like this. And I think it's a huge, huge win for this device. The fastest way to get from Edinburgh to London is generally to take a direct flight which takes about 1h30. However, the other thing it can do, just like its other smart speakers, is to be a speaker and it can broadcast music, radio or notifications on it. The speaker is usable. It is not extremely loud and it can be a little distorted at maximum volume, but lower it by a notch and it's good. It will never be incredible to be such a small device, but it doesn't bother me for occasional listening. One of its advantages is its 3.5mm audio jack, which allows you to take the voice PE and plug it into any other speaker that you could have in your house. So you have a configuration of speakers that click. You can take that, plug it in, and suddenly you have a configuration of smart speakers that click, which is great. And that gives it a big advantage in this area compared to other smart speakers. One thing I would like to see, however, that it doesn't do yet, is that music reading is easier. So I would like to be able to use my voice to read a song via Spotify, which it doesn't do natively. I could probably get around this problem by adding an automation in Home Assistant to manage this, but it's a little more delicate for beginners. And if they could find a way to do this natively on the voice PE without needing an automation, it would be really, really revolutionary. You can't access the Spotify app on your phone, for example, and broadcast it on the voice PE for the moment. It would also be an excellent addition to somehow complete the music reading capabilities on this speaker. However, I hope that we will see great improvements in this area over time thanks to the updates. One thing that constitutes a huge improvement on the PE compared to previous speakers is the voice command during audio reading. So, you know if you've ever tried to use your voice while the music is playing in the room, even in the background, it was practically unusable because the microphone couldn't hear you over the music and it was a little confused because there is suddenly two voices or others and it can't somehow catch what you're saying. But thanks to the audio processor of the voice PE, if you listen to music via the PE and you pronounce the activation mode, it is able to cancel the music it broadcasts and hear you pretty well, sometimes even at a fairly long distance. It also does things like ducking audio. So, if the music is playing and you activate the wake-up mode, the volume decreases and then increases again once the reading is finished. A feature that many people use on smart speakers is the timer. And these are taken care of here on the PE. And when you define a timer, it uses the LED notification ring as a visual indicator of time remaining with a small impulse animation. There are actually a lot of nice little animations that it does with the LED ring, which I love. And it also offers several timers at the same time, which is really, really cool. Minuterie réglée sur 15 secondes. Ok, Nabu. Régler une minuterie sur 15 secondes. La seule chose que j'aimerais voir serait un son de minuterie sélectionnable afin que vous puissiez personnaliser le son que vous obtenez lorsque la minuterie s'active, mais c'est le genre de choses que je suis sûr que nous verrons dans les futures mises à jour. Je n'ai pas vraiment rencontré beaucoup de bugs pour être honnête avec le PE, à part par exemple, si de la musique joue et que vous appuyez sur le bouton du milieu, cela met en pause ce qui est en cours de lecture sur l'enceinte, ce qui est bien. Mais si vous appuyez à nouveau dessus, je m'attendais à ce qu'il reprenne la lecture, mais à la place, il active l'assistance, ce que je peux en quelque sorte comprendre, mais peut-être qu'une option intéressante serait d'avoir un menu déroulant sélectionnable dans les paramètres où vous pouvez en quelque sorte mapper les boutons à l'action que vous voulez qu'il fasse. Cependant, il est facile de résoudre ce problème avec une simple automatisation ou une double pression sur le bouton central du milieu reprend la lecture de ce qui se trouve sur les haut-parleurs. Donc, très facile à personnaliser grâce à la puissance de Home Assistant. Mais pour les nouveaux arrivants, un simple menu déroulant serait super facile. Il y a aussi un son d'œuvre de pack que vous pouvez déclencher si vous pouvez trouver la séquence, que je vous laisse découvrir. Enfin, parlons du nom. Qu'est-ce qu'est en fait une édition d'aperçu ? Et pourquoi s'appelle-t-elle édition d'aperçu ? Et fondamentalement, c'est pour définir des attentes réalistes sur ce qu'est cet appareil. Le nom de l'édition prévue est essentiellement dû au fait qu'ils veulent que les gens comprennent que la voie de Home Assistant est encore en cours de développement et que c'est une expérience assez géniale si vous parlez anglais comme moi. Mais pour certaines des autres langues, celles-ci sont moins complètes pour le moment. Et c'est pour vous faire savoir que votre langue n'est peut-être pas encore tout à fait au point. Mais si vous y réfléchissez, ce matériel est vraiment important même pour ces autres langues, car les gens ont évidemment besoin de matériel pour pouvoir tester et découvrir où se trouvent les lacunes dans les langues et à quel point elles fonctionnent bien. C'est exactement à cela que sert l'édition P.E. Vocale. C'est pour les personnes qui veulent continuer à développer et à améliorer la voix dans Home Assistant tout en ayant une expérience vocale sympa. Maintenant, si vous voulez vérifier la progression de votre langue et même aider, il existe une petite page web sur laquelle vous pouvez vous rendre et qui vous montre toutes les langues sur lesquelles il travaille et à quel point elles sont bien prises en charge. Comme vous pouvez le voir, beaucoup de vert et de bleu, mais il y a aussi certainement du rouge. Alors, assurez-vous de vérifier cela pour vous assurer que vous êtes informé de l'évolution de votre langue. Naturellement, le prix amènera les gens à se demander pourquoi c'est plus cher que les offres d'Amazon ou de Google. Ce n'est pas énormément plus cher que cela, mais mon public, vous êtes assez conscient de ce qui se passe avec ces appareils, mais il est bien connu qu'Amazon, par exemple, fait de grosses pertes sur son haut-parleur intelligent du côté du matériel dans l'espoir de compenser avec le visa que vous leur donnez lorsque vous l'utilisez, eux pour subventionner en quelque sorte le coût. Mais même maintenant, on parle beaucoup d'Amazon devenant un abonnement payant pouvant atteindre 10$ par mois, ce qui dépasserait rapidement le prix d'un PE Vocale. Mais cela mis à part, pour ma part, et je sais que beaucoup d'entre vous ressentent la même chose, je suis prêt à payer plus pour du matériel axé sur la confidentialité et la localisation. Et je pense en fait que le prix est excellent étant donné qu'il remplit cette mission. Je ne dis pas ça seulement parce que nous le vendons. En fait, je m'attendais vraiment à ce que cela coûte un peu plus cher lorsque j'ai entendu que cela allait arriver avant qu'il ne soit annoncé. Et j'ai dit à plusieurs reprises dans le passé que j'étais prêt à payer plus pour du matériel contrôlé localement, ce qui est le cas. Mais encore une fois, n'hésitez pas à consulter certains des autres critiques qui lancent tous des vidéos aujourd'hui pour voir ce qu'ils pensent et pour avoir une image complète du PE Vocale. Pour moi, c'est un très, très bon début et une tentative de matériel vocal sur Home Assistant. Et je suis vraiment excité de voir les gens commencer à l'utiliser, ce qui à son tour va rendre la voix encore meilleure. Et je pense que c'est un peu comme si cet effet allait commencer à faire boule de neige à mesure que les gens commenceront à l'utiliser et à le tester. Et bien sûr, vous ne devriez jamais acheter des choses en fonction des futures mises à jour. Mais je pense que connaissant les antécédents de Home Assistant, cela ne fera que continuer à s'améliorer au cours des prochains temps. Si vous voulez acheter un PE Vocale, je mettrai les liens dans la description. Bien sûr, si vous achetez chez nous, alors c'est génial. Merci beaucoup de nous soutenir. Et si vous souhaitez acheter chez d'autres revendeurs, c'est cool aussi. Je mettrai des liens dans la description où vous trouverez tout. Et c'est à peu près tout pour cette vidéo sur le Voice PE. J'espère vraiment que vous l'avez appréciée. J'aimerais entendre vos réflexions dans les commentaires. J'ai vu beaucoup de vos commentaires au cours des deux dernières années disant, l'année de la voix s'est terminée et il n'y avait plus de matériel. La voix est morte. Et des trucs comme ça. Donc, j'aimerais vraiment entendre ce que vous pensez de cela et si vous prévoyez d'en acheter un, notez dans les commentaires. Merci beaucoup d'avoir regardé. Passez un Noël absolument fantastique et de joyeuses fêtes de ma part et je vous verrai dans la prochaine vidéo.