I'm glad you see how we keep it real on this channel. That is my number one goal to be unbiased in all my videos and tell things like it is. It isn't the best stance for views and clickbait stuff but its the type of videos I would want to see. I saw a few others on this same subject from bigger channels already. Goes to show you though the big issue I have with YT vids. Once you get into it as you require it for revenue and to pay your bills, you have to fake everything and say it is all the best ever stuff, you have to take any paying contract that will come your way and basically be a sell out. You can't be real and tell it like it is. I enjoy the aspect of being able to just say it like it is and have zero plans to change. Looking forward to future development on the hardware and one day see it used full time in my household.
Your honest and unbiased approach is what makes me and other viewers come to your channel, it takes b@lls to be upfront tell the truth and confront the manufacturer's and that's what you have! Keep up the good work pal.
The entity names and getting them to map via voice fluidly in HA has been a problem for a while, even with raspy. They need to make that part smoother and more natural language wise. When I gave it a solid try a while back I gave up on using the built in intents system and went to a big Node Red spider map, but it was absolutely massive and still had limitations.
haha my wife always waits until alexa lights up before she starts talking and I keep telling her she doesn't have to wait. Maybe this device is perfect for her.
I bought mine. Even though it's not the best voice solution it's the only one that respects the user's privacy and also the only one that will work offline. I run home assistant on beefy hardware that handles the voice stuff fairly well. Also smart speakers like Google home and Alexa are subsidize because they gather a lot of information about you. I appreciate your honest review as always!! ❤️
I’ve been using HomeKit with HomePods for a while which is (supposedly) all processed locally and doesn’t track requests. The out of the box Home Assistant solution really isn’t there yet; I ended up building my own voice interface using more powerful transcription and language models which is better, but still not fully complete.
We did try that but it didn't seem to make much of a difference, still a lot of can't understand messages or failed to wake. Maybe things will get better on it being able to process all at once in future updates I hope.
I'd love to see a review comparing Voice PE to the voice assistant Satellite1 just released by Future Proof Homes. Satellite1 includes additional sensors and I like the option to add a decent 25W speaker.
'Preview Edition' I think the clue is in the name. This is for experimenting with and to allow development. At this point it is not a replacement for Google or Alexa. What it leads to is the exciting bit!
I just dont want google mics in my house. 🤷♂️ Would be interested in seeing how it performs with a local llama. I dont mind paying for local processing if it is good.
@digiblurDIY true, but if they get that sorted, its a pretty good option. I agree it’s basically an expensive mic, but I would love to have locally processed home automation controlled by voice. Thanks for the review! Enjoy your holidays ❤️
I have just ordered one despite this review, partly to simply play around with, partly to help progress HA Voice development. Hopefully HA Voice will mature at about the same time that the Big Boys attempt to monetise voice.
You’re the few reviewers who indeed show the reality of the device state instead of hyping it up like others; as users who want to pay for this kind of things, we deserve a real state of things so we can decide if we want it as it is or not
I tell it like it is in my reviews as always. No paid influence stuff to get in the way. I have had some vendors hate it but some appreciate it and correct their mistakes. Those are the stand up ones! This box is a good start but definitely needs some work before it can be used. Only one direction to go from here.
Loved the review. What it showed me was that google is better than my Alexa’s! Never been able to get them to control my smart stuff properly. Played with the other HA voice assistants linked to ChatGPT and they are slow but do a much better job of controlling devices without getting the name perfectly matched. Maybe the new device works better with cloud LLMs but that sort of negates the point. Anyway buying one as it’s geeky tech 😀
I do have an Alexa or two around but we mostly use the Google stuff for the easy casting, google photos, YT music stuff. It is pretty impressive the stuff it understands at times. They can go through phases of dumbing stuff down but then the devs must seem to fix or reverse something at times. Like for a week a few months ago it had no idea what the word shades were.
A lot of comments go over the way it’s presented in this video, except that is not the point. The point is that it’s not as good as a Google Home or Echo. It needs to be matured. Shit happens and it’s the first release. It will be better in the future.
Hi I need help to fix my smart wifi RGB ceiling lights which stuck on warm colour and won't goes to setup mode after couple of attempts as I checked it has WB3L wifi module. Please reply
Does your wife have a You Tube channel for doing reviews? Would so subscribe to that lol. Direct and to the point is awesome. Thanks for the honesty in the review!
The Nest Audios are great. I have them as a pair in my office so it turns into Left and Right channels when I stream music. Nice little punch for those things.
I'm big on automations myself but sometimes I need to change a couple minor things here and there and forget where I put my phone for the 50th time of the day.
I have a couple of Wyoming satellites, which are the same as this thing (cost me like 35-40 dls to build each): really dumb, and the wake work even fires itself; from what I've seen you need to add an LLM for the assistant to really work and understand, but there's 2 camps for that: either use ChatGPT or OLlama, so either you send your data to ooenAI or you use a beefy GPU The cool thing though is that this one has a media player integrated, while on the Wyoming sat you need to install snapcast
The rabbit hole starts to get deep for the typical user running HA there. Many simply do not have the hardware to run that combined with many do not want the power costs to run all that locally.
@Lusanagi my friend just bought one earlier off Seeed after they sent their email out. He's definitely in the vast sea of average Joe users. Not sure what to think or say at this point.
To be fair to the hardware itself, its definitely not the fault of the hardware but probably more so how voice intents are handled in HA by default. I think this is something AI could probably help with.
Thanks for making this honest video. I've been waiting for a good speaker since "the year of the voice" and immediately bought one yesterday when they went to sale. Disappointed with the delay after the wake word, especially since during the official Nabu presentation, they talked up how they have a dedicated chip to handle all the wake word detection and audio filtering. Don't understand why it can't be "always listening" to make wake word faster, especially if everything is open source and we know what it's doing and not transmitting data back to Nabu or other places. WAF above all.
Yes, we've used Google and Alexa speakers for quite some time and never paused with them, you just spoke to it one sentence so it was hard to use this one. It could be some limitation of the chip to not listen all the time or maybe try to buffer to go back? Much like you see camera triggers work now where they detect motion yet they can go back and record the previous 10 seconds before motion hit due to the buffer.
Intro seems a bit harsh. It's clearly not set up with an AI fallback, else it would do all the things the Google device does....and it doesn’t harvest your data lol😊. However, I get the messaging. For those of us who like to tinker it's a good start.
@digiblurDIY Is the wake word just a Google thing? We have a house full of Amazon Echos and we always pause after the wake word 🤷♂️. To be fair, I've never actually tried not pausing 😆. To us, pausing is the norm with a smart speaker.
@@digiblurDIY Paulus and the team said this hardware is for developers working on improving voice and is not intended for "main stream" users, which is why it's called "Preview Edition". The website actually says the device is to "develop this community-driven and private voice assistant".
Is it not being sold to users on websites for HA? The Home Assistant Voice Preview Edition, is the first voice assistant built to seamlessly integrate with Home Assistant. This open source, privacy-focused device has advanced audio processing all contained in a premium design that blends into the home. Holiday Special Offers on Home Assistant Voice! Check the features section of this page for details.
@@digiblurDIY I can't find that quote anywhere on the Home Assistant website, but maybe I am bad at looking. On the product page it says "We’re launching Home Assistant Voice Preview Edition to help accelerate our goal of not only matching the capabilities of existing voice assistants but surpassing them". There is also a section on the product page called "Why Preview Edition" that explains this is not intended for mainstream users. I understand you got your hardware before these pages were up, so I understand your POV, especially if Nabu Casa was not clear in their communication, but the product page is very clear that this is not intended for mainstream users.
My point isn't that you are wrong about the way you use it or that you should go easy on it, but that you are presenting that this is not ready for mainstream without also presenting that that is the intention. This is a "Preview" product and if you are not a developer you should not buy it at this stage.
The way you talk to Nabu, is the same way you needed to talk to Google on older phones. It either is programmed to wait for the word before it listens or its simply too slow to follow (like older phones cant keep up).
My problem with voice is it takes six seconds to process a command. If they can get that down to a couple of seconds without having to buy a quantum conouter I'd be all in.
I'm guessing you'll need some big hardware. I tried the others on the i7-14700k setup and it seemed just okay, but I didn't have the big gaming GPU to run the LLM stuff
Most of the time is taken by speech to text which especially with Whisper is lackluster at accuracy, try wyoming-rhasspy-speech addon instead, it is both fast and much more accurate
I appreciate the honest review. I realize this is an alfa/bata product… but that part aside even after they get most the bugs worked out of it, the hardware and use case for this type of off the shelf product, isn’t worth more than $29.99 & thats being generous in my opinion.
Even if it's true you are also supporting an open source company and it's developers. I pay nabu casa's subscription because I want to support the home assistant project not because they offer something that i can't do myself
@ I agree supporting devs is good. My thought is if it wasn’t the home assessment devs selling this it’d only be worth $10. At $30 2/3would go to the devs. Since this is a product that has use cases where multiples in a system actually makes the product needs to be priced to entice people to want to purchase 6 or 12. They’ll sell more and make more in the long run. But at $60 a pop people may by 1 to support the dev. But theres no way in hell any one will purchase them in quantity at $60 when theres tons of less expensive options with the same ability. In my opinion they’re looking at the short game not the long game.
@mumbles1justin I bought 3. Two for myself and another one as a gift for a friend for the holidays. Also prices go down with scale and distribution. Since this was a fairly small production it was more expensive. FYI they were sold out everywhere in minutes so I guess a lot of people feel the price is worth it
It's even in the name (preview edition). I don't evpect it to work flawlessly out of the box. What I do appreciate with home assistant is that they are constantly making their product better where as GH have been skating along without adding any value to their offering and improving voice recognition. Thanks for the transparency though..I've seen reviewer's edit their videos to cut down on response time. For me ~1-2 sec response time is fine, then again I run my home assistant on higher spec hardware.
Yes I left it unedited in the beginning so you could see the time. Later down the video it got a little much so edited some parts once you saw it. All using the Home Assistant cloud.
Thanks for seeing the real world use reviews we do. Something we take very seriously. I didn't want to stop people from having fun but wanted them to prepare for what they would be getting into.
Digi ya'll are the best! Always keepin' it real.... Lovin' the "Southern thing!" Down south here down under we also string our words and sentences together as well - except it's just one long monotone... so If it won't understand you guys it certainly won't understand me!
I don't know if it has a built in recording buffer, but it really needs one. People are too weirded out by the device 'recording all the time' but the alternative is this goofy wake word activated mic that has to stream back and forth to the HA instance for processing. I'd much rather it back track and send the last X seconds and get stuff done.
I would not usually watch a video where I can't understand what's being said. Is it tested with a neutral voice later in the video? If so I'll skip to that point as comments suggest it's a good review 👍
From up north. Lived on the east coast and down south. They don’t have strong accents at all. Maybe a small drawl but not anything that should cause AI assistants to have difficulty.
@Markyboy767 I should get one of those guys from Swamp People where they have to put his words on the TV to try this device out. But you are right, if we were on the show they wouldn't need the words up.
@digiblurDIY Indeed neither of you speak with a Glaswegian accent. I think I'm just out of practice with American accents, as I don't watch TV these days. A friend pointed out, that another side effect of my lack of TV use, which she has noticed, is that I'm also bad at understanding people who talk faster or fail to enunciate clearly. So unless you are an old-fashioned news reader from the history of the BBC (UK television, not the American type of BBC) speaking the 'Queen's English' I'm probably going to struggle. So I just need to practice 👍 After all, I live in the Netherlands at the moment, and I can no longer detect their Dutch accent.
Wake word processing needs some work then we can get to them handling a bit more with LLM but then that wouldn't be local. We need a cheap Coral type board one day for this.
Even if it worked perfectly for voice commands, which it doesn't, am I the only one disappointed with the choice of built-in speaker? If it is going to be a self-contained unit with built-in audio I'd prefer to have high quality speakers for audio and music playback like Google, Apple, and the Echos. If the speaker is a lousy low end mono speaker, I'd rather have the speaker eliminated, the size reduced, and make it mic only with additional useful automation sensors such as presence sensors with the capability for audio playback on other devices. I know it has the 3.5mm jack, but I don't want multiple speakers all over the house everywhere I want a smart speaker/voice input. I do understand that it is a preview version, but it isn't on par with current tech. Thank you for giving an honest demo of the product without all of the hyperbole. It is certainly not "AMAZING" at this stage, but an early prototype that needs improvement. Kudos to the team for their efforts. It has a lot of potential.
I know the AMAZING comment you are talking about. Lol. We have been discussing on my Discord. Yes. I would have preferred a better speaker. It's like they said hey let's make it sound worse than the Echo dots.
The Satellite1 just launched this morning in what also appears to be an early prototype version. I would love to see you do a comparison of their product. I have no affiliation with them. I like what I’m seeing so far.
Would have been nice and sort of agree, but you don't HAVE to use the built in speaker. It's there if you want it. Ultimately no matter how good they made the speaker, I'd still hate it and want to use a speaker that's good at being a speaker made by a speaker company instead of a speaker that's almost maybe passable at pretending to be a speaker.
that new low power Nvidia AI board thing is promising. it would be nice to have that device loaded with HAOS and use its AI power for this stuff. Assistant on HA is too dumb to be called assistant to be honest.
It's perfectly OK to compare the experiences between Google and HA. Until one realizes why it is you can say the sentence with the wake word together and it will still understand... Right? :)
and for the cost of a google mini, they're going to have to really add some cool features that don't outright block the google integrations. I think the biggest draw could be a solution if your network goes down because google refuses to work if your wifi isn't working, whilst Nabu could perhaps fill that gap.
Luckily I've kicked the cable ISP to the curb after the telco ran fiber so that's a thing of the past. I'm looking forward to some low cost, power efficient methods to run some LLM stuff in the future to offload it all.
How many people are watching this? you expect (and even tell!) that many people to get up and go unplug their smart devices becase YOU WERE TOO LAZY to edit out the wake words..... glad i'm not too lazy to unsub.
Thanks for keeping it real man. Most of the other videos on this speaker are all hype hype hype. "YOU NEED THIS!" They remind me of that Nvidia review asking if you wanted to die without experiencing ray tracing.
Your mistake is integrating with the built-in "conversation engine," which is complete trash. My DIY voice system connected to local LLM works MUCH better than either of these.
Nabu is no where near good enough to be even a manageable alternative, and it utterly disappointing because A- you chose the OLDEST google home device to pit it against along with the rotted state of google assistant and it still failed hard and B- given said state of google assistant i was hoping that this would be a jumping off point...
The “year of the voice” was a flop. Making it worse, the world discovered LLM’s as chat GPT blew up during that same year. People become instantly accustomed to what that new tech can do, instantly rendering their hard work nearly obsolete. Many of us wish Alexa and Google had synthetic neurons making them smarter which both have been actively working on (likely with a cost). Those people certainly don’t want to use a device (local or not) that’s a worse user experience. Local models are great but like many, my hardware struggles. The integrations just isn’t there yet and the models everyone chooses to specifically train are usually genetic, ‘fully aligned’ llama models for HA which are too censored for my taste. They also struggle with HA in comparison to OAI. I want my assistant running on totally unrestrained and uncensored models, not only comprehending my basic requests but I want it to speak like my family or I do and use “naughty” or “four letter” words, be funny and even ultra sarcastic when I feel like it. I want my assistant to ridicule me for asking a basic math question or bust my balls by calling me a lazy aśś when I ask it to turn on the fan or dryer. I want it to say things like “you better clean that ass” when I ask it to turn in the shower. If, I Initiate the “romantic” scene I want it to make a joke about getting lucky and when I change the scene back, I want it to criticize me for “being too quick” . I also want my assistant to be able to help my family set up quick and easy automations automagically. They should be able to say “hey [wake word] can you set the kitchen light to come on at 4PM each night except weekends” with follow up questions that are necessary to build a good automation. I want my assistant to automatically notice that my coffee pot is still on because the safety timer automation was accidentally turned off. To recognize areas or understand presence. If I say turn of the fan while in the bathroom it should know I’m in the bathroom then respond with “are you sure? It smells like shιτ in here” while it executes the command on the proper fan. I want it to seriously question someone if they ask for something, stupid, strange or dangerous like setting the thermostat to 90°, turning the washer on after it just stoped, or using amperage nearing a breakers capacity if someone recently requested a space heater to come on. I want it to recognize that window is opened or left opened when it’s 18° outside without writing automations. If i ask my assistant to talk dirty to me or to tell me how to make crack, it better sound like a Jenna Jameson in her prime or have me grabbing the pot and baking soda as I’m an adult and don’t need PG 13 restrictions. Life is too boring. If I’m feeling insecure, I want my assistant to tell me that I’m so sexy it could just *** **** ***** me, to make me feel better about myself. Then when my wife interacts with it, it lets her know I’m feeling down and she should “cheer me up”. Extra Hyperbolic example but still….y’all get the picture. I’m my mother in laws phone is detected on the network I want it to say brace yourself Susie’s here. If I say “add larger jeans to the shopping list” , I want my assistant to ask if I’m just a fat aśś eating too much or if my package is finally growing! Paulus and the HA/NABU team have always pushed back against LLM’s and AI. Years ago he even said that they didn’t see any worthy use case for HA. They recently tried playing the environmental card to refute its use. I strongly believe that position stems from all their time and investment during the “year of the voice” or (‘multiple’ years of the voice) and what they had originally envisioned.
> Paulus and the HA/NABU team have always pushed back against LLM’s and AI. No, there was pushing back against using chatgpt to generate answers to questions in the community forum and it is still not allowed. There was never push back against using LLMs in HA and there's plenty of choices for that now. The rest of your list is simply describing the potential of using llms with an assistant. You aren't going to get this from Google or Amazon. You might get it with a locally running llm that yes, you're going to need to run and invest in.
I couldn't agree less with your statements. I think the year of the voice was great! Is it perfect absolutely not but it's miles ahead from where it was and having an open source solution is amazing. It's not perfect for sure and I'm sure it will only get better over time
![Seungmin "그렇게, 천천히, 우리(As we are)" | [Stray Kids : SKZ-PLAYER]](http://i.ytimg.com/vi/kAzmhLHePqU/mqdefault.jpg)
I really appreciate not making it look better than it is. I am very excited for this, I had a feeling the first iteration would need some work.
I'm glad you see how we keep it real on this channel. That is my number one goal to be unbiased in all my videos and tell things like it is. It isn't the best stance for views and clickbait stuff but its the type of videos I would want to see.
I saw a few others on this same subject from bigger channels already. Goes to show you though the big issue I have with YT vids. Once you get into it as you require it for revenue and to pay your bills, you have to fake everything and say it is all the best ever stuff, you have to take any paying contract that will come your way and basically be a sell out. You can't be real and tell it like it is.
I enjoy the aspect of being able to just say it like it is and have zero plans to change.
Looking forward to future development on the hardware and one day see it used full time in my household.
Your honest and unbiased approach is what makes me and other viewers come to your channel, it takes b@lls to be upfront tell the truth and confront the manufacturer's and that's what you have! Keep up the good work pal.
I'm also excited to get one. Damn the WAF! We'll just live dual voice assistant household for now!
The entity names and getting them to map via voice fluidly in HA has been a problem for a while, even with raspy. They need to make that part smoother and more natural language wise. When I gave it a solid try a while back I gave up on using the built in intents system and went to a big Node Red spider map, but it was absolutely massive and still had limitations.
haha my wife always waits until alexa lights up before she starts talking and I keep telling her she doesn't have to wait. Maybe this device is perfect for her.
They're just lucky Travis wouldn't let me throw that thing from frustration while recording "he still needed it".
I bought mine.
Even though it's not the best voice solution it's the only one that respects the user's privacy and also the only one that will work offline.
I run home assistant on beefy hardware that handles the voice stuff fairly well.
Also smart speakers like Google home and Alexa are subsidize because they gather a lot of information about you.
I appreciate your honest review as always!! ❤️
Yeah, I am not looking forward to the day when I have to choose between the privacy of my data and having something actually work.
I’ve been using HomeKit with HomePods for a while which is (supposedly) all processed locally and doesn’t track requests. The out of the box Home Assistant solution really isn’t there yet; I ended up building my own voice interface using more powerful transcription and language models which is better, but still not fully complete.
Wow, who knew there's a voice assistant worse than Siri.
I have heard that one is pretty awful
If you turn off the wake word sound/beep you don't need to pause after "Okay Nabu", just like with the Google device.
We did try that but it didn't seem to make much of a difference, still a lot of can't understand messages or failed to wake. Maybe things will get better on it being able to process all at once in future updates I hope.
@@digiblurDIYtry using wyoming-rhasspy-speech add-on for speech to text, it helped me a lot with recognition
I'd love to see a review comparing Voice PE to the voice assistant Satellite1 just released by Future Proof Homes. Satellite1 includes additional sensors and I like the option to add a decent 25W speaker.
I ordered one that night they launched. Time to persuade the wife again.
Shows how much money Google and Amazon poured into getting their voice assistants working so well so early on.
Very true. The Google speaker I used was from 2016.
Right, but customers don't care. They want it to work.
Another great video! Maybe the Nabu does go well on the U7Pro. 😂
I made sure it wasn't using the U7 AP during this test otherwise it would have been twice as bad.
Yep, I bought one. Mostly as a fast way to get into this voice assistant but I don't expect to use it.
Voice assistants never caught on in my home.
Be interesting to see how it evolves
Travis, Nice Pear! ;-) BTW,... Mrs. Digi's look says it all! Thanks for the Video review; I will hold off (for now).
They are delicious!
'Preview Edition' I think the clue is in the name. This is for experimenting with and to allow development. At this point it is not a replacement for Google or Alexa. What it leads to is the exciting bit!
That is the name indeed but this was a full on launch product sold in mass quantities in stores.
Yeah agree with not making it look better than it is. Your videos are just so good, thank you!
Thank you! I always strive to do some real world reviews.
I just dont want google mics in my house. 🤷♂️
Would be interested in seeing how it performs with a local llama. I dont mind paying for local processing if it is good.
I think the bigger issue is the waking up and then trying to process it. LLM kinda comes third on that part once they get step 1 and 2 done.
@digiblurDIY true, but if they get that sorted, its a pretty good option. I agree it’s basically an expensive mic, but I would love to have locally processed home automation controlled by voice.
Thanks for the review! Enjoy your holidays ❤️
Yep. Definitely will keep my eye on future updates of it.
I have just ordered one despite this review, partly to simply play around with, partly to help progress HA Voice development. Hopefully HA Voice will mature at about the same time that the Big Boys attempt to monetise voice.
Nothing wrong with that. Still cool to tinker with.
You’re the few reviewers who indeed show the reality of the device state instead of hyping it up like others; as users who want to pay for this kind of things, we deserve a real state of things so we can decide if we want it as it is or not
I tell it like it is in my reviews as always. No paid influence stuff to get in the way. I have had some vendors hate it but some appreciate it and correct their mistakes. Those are the stand up ones!
This box is a good start but definitely needs some work before it can be used. Only one direction to go from here.
what motorized blinds do you use?
I'm using Smartwings 100% blackout Zigbee roller shades - amzn.to/41Kl7Nb
Her accent messes up the Nabu wake word XD
Nabu has the accent
Loved the review. What it showed me was that google is better than my Alexa’s! Never been able to get them to control my smart stuff properly. Played with the other HA voice assistants linked to ChatGPT and they are slow but do a much better job of controlling devices without getting the name perfectly matched. Maybe the new device works better with cloud LLMs but that sort of negates the point.
Anyway buying one as it’s geeky tech 😀
I do have an Alexa or two around but we mostly use the Google stuff for the easy casting, google photos, YT music stuff. It is pretty impressive the stuff it understands at times. They can go through phases of dumbing stuff down but then the devs must seem to fix or reverse something at times. Like for a week a few months ago it had no idea what the word shades were.
does the google home with HA work while offline/locally?
I haven't tried since they did the new offline matter release.
Gotta love Travis and his honest reviews. No one does it better!
Woot woot!
it would be nice if you also made an ai ollama server to answer to all your requests
if you use llama 3.1 or 3.2 it understands the context and execute the action correctly, its a lot of work getting it all running, butt so worth it
If it can run on an N100 then let's roll.
You in Baton Rouge? I'm in the lafayette area LOL.
What up neighbor!?
A lot of comments go over the way it’s presented in this video, except that is not the point. The point is that it’s not as good as a Google Home or Echo. It needs to be matured. Shit happens and it’s the first release. It will be better in the future.
Yup. Cool product for initial usage. But not ready yet. They had to start somewhere but I feel the launch was a bit too wide in the audience
Hi I need help to fix my smart wifi RGB ceiling lights which stuck on warm colour and won't goes to setup mode after couple of attempts as I checked it has WB3L wifi module. Please reply
Esphome?
Does your wife have a You Tube channel for doing reviews? Would so subscribe to that lol. Direct and to the point is awesome. Thanks for the honesty in the review!
Hahah! She actually has taken an interest in doing some more shorts for products on this channel and will be doing more regular videos.
How do you get Google next speakers on a discount!? I would like to try as well
Showing $50 on their store here and you get money back too if you have Google One. store.google.com/product/nest_audio?hl=en-US
@digiblurDIY I had no idea and been a Google one subscriber for years! Thank you!
The Nest Audios are great. I have them as a pair in my office so it turns into Left and Right channels when I stream music. Nice little punch for those things.
@@digiblurDIY looking to do something similar. Do they experience voice delay if I have them as a pair for my Chromecast?
I haven't had that issue. They act like normal. The pair comes up as a speaker group/device.
I've only used the phone app a few times for voice. I can find most entities faster than I could even ask Google anyway.
I'm big on automations myself but sometimes I need to change a couple minor things here and there and forget where I put my phone for the 50th time of the day.
@digiblurDIY that can be an issue for sure.
I have a couple of Wyoming satellites, which are the same as this thing (cost me like 35-40 dls to build each): really dumb, and the wake work even fires itself; from what I've seen you need to add an LLM for the assistant to really work and understand, but there's 2 camps for that: either use ChatGPT or OLlama, so either you send your data to ooenAI or you use a beefy GPU
The cool thing though is that this one has a media player integrated, while on the Wyoming sat you need to install snapcast
The rabbit hole starts to get deep for the typical user running HA there. Many simply do not have the hardware to run that combined with many do not want the power costs to run all that locally.
@digiblurDIY yeah.... definitely the whole voice assistant thing is niche and not ready for the vast majority of users
@Lusanagi my friend just bought one earlier off Seeed after they sent their email out. He's definitely in the vast sea of average Joe users. Not sure what to think or say at this point.
Mine works great but unfortunately I don't have a yapping wife to test it out. 😁
Turn on the TV
Thank you beta testers. It’ll be ready next year. I’ll purchase mine then.
To be fair to the hardware itself, its definitely not the fault of the hardware but probably more so how voice intents are handled in HA by default. I think this is something AI could probably help with.
Yep. I think it can get there with some fixes on the Home Assistant cloud side and better wake word processing.
Thanks for making this honest video. I've been waiting for a good speaker since "the year of the voice" and immediately bought one yesterday when they went to sale. Disappointed with the delay after the wake word, especially since during the official Nabu presentation, they talked up how they have a dedicated chip to handle all the wake word detection and audio filtering.
Don't understand why it can't be "always listening" to make wake word faster, especially if everything is open source and we know what it's doing and not transmitting data back to Nabu or other places. WAF above all.
Yes, we've used Google and Alexa speakers for quite some time and never paused with them, you just spoke to it one sentence so it was hard to use this one. It could be some limitation of the chip to not listen all the time or maybe try to buffer to go back? Much like you see camera triggers work now where they detect motion yet they can go back and record the previous 10 seconds before motion hit due to the buffer.
If you set your configuration (ready with speechdetection) not to default but to agressive the reactions are faster
I actually had more issues with it on aggressive. Default seemed to work best. Well somewhat
Intro seems a bit harsh. It's clearly not set up with an AI fallback, else it would do all the things the Google device does....and it doesn’t harvest your data lol😊. However, I get the messaging. For those of us who like to tinker it's a good start.
AI fallback isn't going to fix the wake word though and understanding regular commands. We can get to the who is the president questions later.
That is the biggest issue with this review. Nobody expects this should work without AI fallback.
@digiblurDIY Is the wake word just a Google thing? We have a house full of Amazon Echos and we always pause after the wake word 🤷♂️. To be fair, I've never actually tried not pausing 😆. To us, pausing is the norm with a smart speaker.
Would AI fallback fix the wake word and lack of understanding things?
I have a couple Alexas still too and we don't pause using those either. We never have.
You should have used AI fallback it would have answered stuff like weather
I set it up with the HA Cloud like the main stream users would have using Nabu Casa on their Green, Yellow, Pi, etc.
@@digiblurDIY Paulus and the team said this hardware is for developers working on improving voice and is not intended for "main stream" users, which is why it's called "Preview Edition". The website actually says the device is to "develop this community-driven and private voice assistant".
Is it not being sold to users on websites for HA?
The Home Assistant Voice Preview Edition, is the first voice assistant built to seamlessly integrate with Home Assistant. This open source, privacy-focused device has advanced audio processing all contained in a premium design that blends into the home.
Holiday Special Offers on Home Assistant Voice! Check the features section of this page for details.
@@digiblurDIY I can't find that quote anywhere on the Home Assistant website, but maybe I am bad at looking.
On the product page it says "We’re launching Home Assistant Voice Preview Edition to help accelerate our goal of not only matching the capabilities of existing voice assistants but surpassing them".
There is also a section on the product page called "Why Preview Edition" that explains this is not intended for mainstream users.
I understand you got your hardware before these pages were up, so I understand your POV, especially if Nabu Casa was not clear in their communication, but the product page is very clear that this is not intended for mainstream users.
My point isn't that you are wrong about the way you use it or that you should go easy on it, but that you are presenting that this is not ready for mainstream without also presenting that that is the intention. This is a "Preview" product and if you are not a developer you should not buy it at this stage.
The way you talk to Nabu, is the same way you needed to talk to Google on older phones. It either is programmed to wait for the word before it listens or its simply too slow to follow (like older phones cant keep up).
My problem with voice is it takes six seconds to process a command. If they can get that down to a couple of seconds without having to buy a quantum conouter I'd be all in.
I'm guessing you'll need some big hardware. I tried the others on the i7-14700k setup and it seemed just okay, but I didn't have the big gaming GPU to run the LLM stuff
@@digiblurDIY Well, damn. I was hoping it would be more usable on my NAS vs a RPi. But, I don't come close to having i7-14700k speed.
@@digiblurDIY you don't need a crazy, powerful GPU to run a llama 3.1 8b instance I mean we can get away with a 3060 12gb
Most of the time is taken by speech to text which especially with Whisper is lackluster at accuracy, try wyoming-rhasspy-speech addon instead, it is both fast and much more accurate
I appreciate the honest review. I realize this is an alfa/bata product… but that part aside even after they get most the bugs worked out of it, the hardware and use case for this type of off the shelf product, isn’t worth more than $29.99 & thats being generous in my opinion.
Even if it's true you are also supporting an open source company and it's developers.
I pay nabu casa's subscription because I want to support the home assistant project not because they offer something that i can't do myself
@ I agree supporting devs is good. My thought is if it wasn’t the home assessment devs selling this it’d only be worth $10. At $30 2/3would go to the devs. Since this is a product that has use cases where multiples in a system actually makes the product needs to be priced to entice people to want to purchase 6 or 12. They’ll sell more and make more in the long run. But at $60 a pop people may by 1 to support the dev. But theres no way in hell any one will purchase them in quantity at $60 when theres tons of less expensive options with the same ability. In my opinion they’re looking at the short game not the long game.
@mumbles1justin I bought 3.
Two for myself and another one as a gift for a friend for the holidays.
Also prices go down with scale and distribution.
Since this was a fairly small production it was more expensive.
FYI they were sold out everywhere in minutes so I guess a lot of people feel the price is worth it
@ Like I said just my opinion
Thank you for showing it how it really is. THIS is the type of review I need. You just gained a subscriber.
Appreciate that! We tell it like it is in our reviews and projects.
hey you need to set it up better, especially whisper
use it at least small. int8, the tiny one is horrible
We used the recommended Nabu Casa method as mentioned since a lot of people have the slower green and Pi setups and no GPU boxes.
It's even in the name (preview edition). I don't evpect it to work flawlessly out of the box. What I do appreciate with home assistant is that they are constantly making their product better where as GH have been skating along without adding any value to their offering and improving voice recognition. Thanks for the transparency though..I've seen reviewer's edit their videos to cut down on response time. For me ~1-2 sec response time is fine, then again I run my home assistant on higher spec hardware.
Yes I left it unedited in the beginning so you could see the time. Later down the video it got a little much so edited some parts once you saw it. All using the Home Assistant cloud.
I really appreciate this review. wish I had seen it before I ordered but it probably would not have impacted my choice to buy it.
Thanks for seeing the real world use reviews we do. Something we take very seriously. I didn't want to stop people from having fun but wanted them to prepare for what they would be getting into.
Digi ya'll are the best! Always keepin' it real.... Lovin' the "Southern thing!" Down south here down under we also string our words and sentences together as well - except it's just one long monotone... so If it won't understand you guys it certainly won't understand me!
Lol... Hoping for a redo after a few months or so of some firmware and software updates
I don't know if it has a built in recording buffer, but it really needs one. People are too weirded out by the device 'recording all the time' but the alternative is this goofy wake word activated mic that has to stream back and forth to the HA instance for processing. I'd much rather it back track and send the last X seconds and get stuff done.
yeah it definitely needs this buffer to go back. Hoping they can add that in future firmware updates.
I have no idea what she is saying 😂 I'm terrible at understanding people with thick (strong) accents
I would not usually watch a video where I can't understand what's being said. Is it tested with a neutral voice later in the video? If so I'll skip to that point as comments suggest it's a good review 👍
What is considered neutral? Neither of us have thick accents.
From up north. Lived on the east coast and down south. They don’t have strong accents at all. Maybe a small drawl but not anything that should cause AI assistants to have difficulty.
@Markyboy767 I should get one of those guys from Swamp People where they have to put his words on the TV to try this device out. But you are right, if we were on the show they wouldn't need the words up.
@digiblurDIY Indeed neither of you speak with a Glaswegian accent. I think I'm just out of practice with American accents, as I don't watch TV these days. A friend pointed out, that another side effect of my lack of TV use, which she has noticed, is that I'm also bad at understanding people who talk faster or fail to enunciate clearly. So unless you are an old-fashioned news reader from the history of the BBC (UK television, not the American type of BBC) speaking the 'Queen's English' I'm probably going to struggle.
So I just need to practice 👍
After all, I live in the Netherlands at the moment, and I can no longer detect their Dutch accent.
Defo regret hype buying it. Thanks for your honesty!
Any time! Hoping to see some improvements and let the family try it again in 3 or 4 months
i feel like LLM will fix most of your issues
Wake word processing needs some work then we can get to them handling a bit more with LLM but then that wouldn't be local. We need a cheap Coral type board one day for this.
Even if it worked perfectly for voice commands, which it doesn't, am I the only one disappointed with the choice of built-in speaker? If it is going to be a self-contained unit with built-in audio I'd prefer to have high quality speakers for audio and music playback like Google, Apple, and the Echos. If the speaker is a lousy low end mono speaker, I'd rather have the speaker eliminated, the size reduced, and make it mic only with additional useful automation sensors such as presence sensors with the capability for audio playback on other devices. I know it has the 3.5mm jack, but I don't want multiple speakers all over the house everywhere I want a smart speaker/voice input. I do understand that it is a preview version, but it isn't on par with current tech. Thank you for giving an honest demo of the product without all of the hyperbole. It is certainly not "AMAZING" at this stage, but an early prototype that needs improvement. Kudos to the team for their efforts. It has a lot of potential.
I know the AMAZING comment you are talking about. Lol. We have been discussing on my Discord.
Yes. I would have preferred a better speaker. It's like they said hey let's make it sound worse than the Echo dots.
The Satellite1 just launched this morning in what also appears to be an early prototype version. I would love to see you do a comparison of their product. I have no affiliation with them. I like what I’m seeing so far.
@SaltydogNC I checked out their page as well and picked one up last night. Thanks for making sure I saw it as well.
Would have been nice and sort of agree, but you don't HAVE to use the built in speaker. It's there if you want it. Ultimately no matter how good they made the speaker, I'd still hate it and want to use a speaker that's good at being a speaker made by a speaker company instead of a speaker that's almost maybe passable at pretending to be a speaker.
that new low power Nvidia AI board thing is promising.
it would be nice to have that device loaded with HAOS and use its AI power for this stuff. Assistant on HA is too dumb to be called assistant to be honest.
I almost pulled the trigger on it but saw it was a bit limited with only 8GB
It's perfectly OK to compare the experiences between Google and HA. Until one realizes why it is you can say the sentence with the wake word together and it will still understand... Right? :)
Exactly! It is hard to change and go backwards.
Is it just me or do voice assistants in general not like women's voices as much as men's voices?
We never noticed that on Google and Alexa.
@digiblurDIY My wife has way more issues with Alexa then me.
I just wish Nabu Casa would focus more on fixing the HA core code before doing pet projects like this. For example security is none existant in HA.
I mean if you're whole argument is "I'm in the Google ecosystem and I don't want to change" then it never had a chance.
Nope. I want to go local with an open source box.
and for the cost of a google mini, they're going to have to really add some cool features that don't outright block the google integrations. I think the biggest draw could be a solution if your network goes down because google refuses to work if your wifi isn't working, whilst Nabu could perhaps fill that gap.
Luckily I've kicked the cable ISP to the curb after the telco ran fiber so that's a thing of the past. I'm looking forward to some low cost, power efficient methods to run some LLM stuff in the future to offload it all.
How many people are watching this? you expect (and even tell!) that many people to get up and go unplug their smart devices becase YOU WERE TOO LAZY to edit out the wake words.....
glad i'm not too lazy to unsub.
I wasn't lazy to edit it out. I tried to and it sounded like crap. I tried other sound effects and it was horrible and childish sounded. See ya!
Thanks for keeping it real man. Most of the other videos on this speaker are all hype hype hype. "YOU NEED THIS!" They remind me of that Nvidia review asking if you wanted to die without experiencing ray tracing.
Definitely! Keeping it real as always.
I hope in future versions it will have a 5 Ghz wifi.
Yes! That would be nice but the esp32-c5 hasn't really shown its head much.
It's the infamous red ring of death!
Your mistake is integrating with the built-in "conversation engine," which is complete trash.
My DIY voice system connected to local LLM works MUCH better than either of these.
I went with the one they recommended.
Not yet ready for prime time.
Nabu fail a lot 😄
It has been pretty entertaining.
Nabu is no where near good enough to be even a manageable alternative, and it utterly disappointing because A- you chose the OLDEST google home device to pit it against along with the rotted state of google assistant and it still failed hard and B- given said state of google assistant i was hoping that this would be a jumping off point...
YUP! That trusty old original Google Home is still alive and doing well!
It obviously needs work. If it was $9.99 I might consider getting one just for testing, but $60 ????
The Seeedstudio Respeaker one is a bit cheaper.
The “year of the voice” was a flop. Making it worse, the world discovered LLM’s as chat GPT blew up during that same year. People become instantly accustomed to what that new tech can do, instantly rendering their hard work nearly obsolete. Many of us wish Alexa and Google had synthetic neurons making them smarter which both have been actively working on (likely with a cost). Those people certainly don’t want to use a device (local or not) that’s a worse user experience. Local models are great but like many, my hardware struggles. The integrations just isn’t there yet and the models everyone chooses to specifically train are usually genetic, ‘fully aligned’ llama models for HA which are too censored for my taste. They also struggle with HA in comparison to OAI. I want my assistant running on totally unrestrained and uncensored models, not only comprehending my basic requests but I want it to speak like my family or I do and use “naughty” or “four letter” words, be funny and even ultra sarcastic when I feel like it. I want my assistant to ridicule me for asking a basic math question or bust my balls by calling me a lazy aśś when I ask it to turn on the fan or dryer. I want it to say things like “you better clean that ass” when I ask it to turn in the shower. If, I Initiate the “romantic” scene I want it to make a joke about getting lucky and when I change the scene back, I want it to criticize me for “being too quick” . I also want my assistant to be able to help my family set up quick and easy automations automagically. They should be able to say “hey [wake word] can you set the kitchen light to come on at 4PM each night except weekends” with follow up questions that are necessary to build a good automation. I want my assistant to automatically notice that my coffee pot is still on because the safety timer automation was accidentally turned off. To recognize areas or understand presence. If I say turn of the fan while in the bathroom it should know I’m in the bathroom then respond with “are you sure? It smells like shιτ in here” while it executes the command on the proper fan. I want it to seriously question someone if they ask for something, stupid, strange or dangerous like setting the thermostat to 90°, turning the washer on after it just stoped, or using amperage nearing a breakers capacity if someone recently requested a space heater to come on. I want it to recognize that window is opened or left opened when it’s 18° outside without writing automations. If i ask my assistant to talk dirty to me or to tell me how to make crack, it better sound like a Jenna Jameson in her prime or have me grabbing the pot and baking soda as I’m an adult and don’t need PG 13 restrictions. Life is too boring. If I’m feeling insecure, I want my assistant to tell me that I’m so sexy it could just *** **** ***** me, to make me feel better about myself. Then when my wife interacts with it, it lets her know I’m feeling down and she should “cheer me up”. Extra Hyperbolic example but still….y’all get the picture. I’m my mother in laws phone is detected on the network I want it to say brace yourself Susie’s here. If I say “add larger jeans to the shopping list” , I want my assistant to ask if I’m just a fat aśś eating too much or if my package is finally growing!
Paulus and the HA/NABU team have always pushed back against LLM’s and AI. Years ago he even said that they didn’t see any worthy use case for HA. They recently tried playing the environmental card to refute its use. I strongly believe that position stems from all their time and investment during the “year of the voice” or (‘multiple’ years of the voice) and what they had originally envisioned.
> Paulus and the HA/NABU team have always pushed back against LLM’s and AI.
No, there was pushing back against using chatgpt to generate answers to questions in the community forum and it is still not allowed. There was never push back against using LLMs in HA and there's plenty of choices for that now.
The rest of your list is simply describing the potential of using llms with an assistant. You aren't going to get this from Google or Amazon. You might get it with a locally running llm that yes, you're going to need to run and invest in.
Feel free to expand their feature set to work as you see fit it is open source after all
I couldn't agree less with your statements.
I think the year of the voice was great! Is it perfect absolutely not but it's miles ahead from where it was and having an open source solution is amazing.
It's not perfect for sure and I'm sure it will only get better over time
Dear, did you receive the email I sent you? Looking forward to our cooperation🥰
---Sikaic Brand Manager
Looks like you haven't responded yet.