just got a weather alert: expect a foot of snow starting in about two hours

#SnowBoy!

Fantastic compilation of #LatinJazz done by Snowboy and friends.....
#Snowboy And The Latin Section – Ritmo Snowbo cd on #AcidJazz 1989
http://youtube.com/playlist?list=OLAK5uy_mrS7UXSJIjBOpfWNNC1vmQ1XZ4U-f5XfM

#ThisAlsoHappened and #NoOneDied; #AtLeast, #NoOne we #KnowOf...

🧙⚔️🤖:wolfparty:​🤖⚔️🧙 | 🪐🦹🚀🦄🚀🦹🪐

#Snowboy: Theme from The New Avengers

https://www.youtube.com/watch?v=nTMMginCMl0

I...I...

...I did it.

I saw my kids do it and I tried myself, but never did I think I'd pull it off.

I made a tiny, adorable, perfect Snowboy.

I'm just speechless and love it so much. And I happened to do it while dressed as Doom Guy.

Perfection.

Have you ever made a tiny Snowboy? Let me know!

#ACNH #AnimalCrossing #VideoGames #Nintendo #NintendoSwitch #ACNHScreenshots #ACNHCommunity #あつもり #あつ森 #CozyGames #CozyGamer #CozyGaming #Snowboy #Snowman #ACNHSnowboy #ACNHWinter

Okay, so my 9 year olds today both wanted to make perfect Snowboys. One had me help them, so I made a standard, big Snowboy.

The other said, "I am making a tiny one!" I had never seen a tiny, perfect Snowboy until this one. It's adorable, so had to share.

#ACNH #AnimalCrossing #VideoGames #Nintendo #NintendoSwitch #ACNHScreenshots #ACNHCommunity #あつもり #あつ森 #CozyGames #CozyGamer #CozyGaming #Snowboy #Snowman #ACNHSnowboy #Parenting #ACNHWinter

My history of making Snowboys in Animal Crossing: New Horizons.

2020 - One perfect, the rest not
2021 - None perfect
2022 - Only made one, then gave up on Snowboys
2023 - Made 3, all perfect

I'm learning. Only took me 4 years of playing, but I'm learning 🤣

#ACNH #AnimalCrossing #VideoGames #Nintendo #NintendoSwitch #ACNHScreenshots #ACNHCommunity #あつもり #あつ森 #CozyGames #CozyGamer #CozyGaming #Snowboy #Snowman #ACNHSnowboy #ACNHSnowman #ACNHScene

Me - Last year: Did not make a single perfect Snowboy
Me - 2 days into in-game winter: made a perfect Snowboy both days.

Finally.

#ACNH #AnimalCrossing #Snowboy #ACNHSnowboy

It's good that other people are also bringing up the elephant in the room: why do you need to pay money for one more electronic gadget that listens to you 24/7, when voice assistants aren't supposed to be rocket science in 2023 anymore? https://news.ycombinator.com/item?id=35857631

I wrote two articles on how to build custom #VoiceAssistants using just a Raspberry Pi and a microphone, one in 2019 https://blog.platypush.tech/article/Build-your-customizable-voice-assistant-with-Platypush and one in 2020 https://blog.platypush.tech/article/Build-custom-voice-assistants.
It's definitely doable and I still have my own custom assistants in the house. However, I had to get around with a #Snowboy model for hotword detection (and Snowboy is now basically abandoned), Mozilla #DeepSpeech model for speech-to-text (and that's quite heavy), and #Mycroft's mimic3 text-to-speech model (and Mycroft is now basically bankrupt). Then writing the integration is relatively easy - I used #Platypush, but it can definitely be done with Home Assistant and OpenHAB too.

Compared to 3-4 years ago, I think we're now in a state where the content is no longer the issue (just plug into a LLM, and all of your text requests will get an answer), nor integrations are a problem (just write a Platypush event hook on speech detected, and you can connect it to everything, no need for "Works with Google/Alexa" labels). Text-to-speech synthesis has also become cheap and ubiquitous.

But the hotword detection and speech-to-text models are still IMHO the bottleneck. Hotword detection is a field where you need a very small and lightweight model that only detects a specific word or phrase in a very reliable way. Snowboy was an amazing FOSS project - which also came with this cool idea of "crowd-funded models", where in order to download a model for a certain hotword you were first supposed to provide three audio tracks where you say that word in order to improve the model. But it's now discontinued because it cost the volunteers too much to run the infra.

And Mozilla DeepSpeech is a relatively good choice for general-purpose speech-to-text, but it's heavy (it takes 100% of the CPU when it runs on a Raspberry Pi) and it's mostly optimized for English - even support for other Western languages is patchy. OpenAI's recent Whisper model seems like a solid alternative, but it's also plagued by the 100% CPU issue - also, I no longer trust anything that comes from OpenAI, no matter how noble some of their efforts may look.

If there are other open-source alternatives that solve these problems, I'd be very happy to learn about them. Once these blockers are removed, there should be really no reason for anyone to feed their audio streams to Google or Amazon.

In the meantime, I'm planning to spend some time playing with some self-hosted LLM model to see if I can replace the Google Assistant library on the last Raspberry Pi that runs it in my home.

@jonty @mattgrayyes #Snowboy to the rescue, perhaps? <Some time later...> "Eh up, Google, me duck" 🦆

https://community.element14.com/challenges-projects/element14-presents/b/blog/posts/how-to-use-a-custom-wake-word-with-google-assistant

@tyil as somebody who has tinkered with voice assistants for a while, I think that we should also keep in mind that this is a field that still requires some trade-offs.

Of course the goal is to have full on-device processing, but that still comes at a cost in terms of resources.

There are probably two models for self-hosted STT that are currently the state-of-the-art - Mozilla's DeepSpeech and OpenAI's Whisper. I've tinkered extensively with both of them, and even though I'm quite impressed by the results (at least for English), they were also keeping the CPU quite warm. Running them on my Dell was still ok-ish, running them on a RPi4 is ok only if you use the device ONLY for that, and running them smoothly on a RPi3 (or anything below) is just not an option. This means that, realistically, a company that wants to ship a fully FLOSS box with fully on-device processing, as of now, would still have to price its box ~$250 just to cover the costs of the hardware - and that's not a very compelling option when people can buy voice assistants at 10% of that price tag.

That's, I think, what has put Mycroft in this uncomfortable position. If they want to ship a fully self-hosted solution, they'd have to price their box well above market prices. Then they went for a "hybrid" approach where they support both on-device and cloud-based backends. But that means that their box, by default, is just an alternative (and more expensive) cloud-based assistant (like Google's or Amazon's), and that's not a compelling selling point. On the other hand, it means that yes, those with the time and skills to tinker can still use the open-source core to build their own self-hosted solution, but those are hard to monetize and don't help attract investment. So, in lack of public funding, you end up with a product that on one hand fails to build a compelling commercial case for an alternative, and on the other hand fails to convince investors that money can be made out of the developer community.

Btw a project that went through a similar fate (and also provided a very interesting solution with good trade-offs) was #Snowboy. It was a STT engine narrowly focused on detecting hotwords, not the whole speech, it was crowd-trained (meaning that in order to download the model for a hotword from their website you first had to upload three audio samples of you saying that hotword), and it was a neat way to split the problem of STT in two:

1. Hotword detection (it uses small models, it's narrowly focused on detecting just the hotword, it's lightweight and efficient, and it probably runs 99% of the time)

2. Speech detection (it uses large models, it can detect nearly anything said in a particular language, but it's heavy and performance-hungry, and, since you only use it 1% of the time, it shouldn't make sense to run it all the time - you only start it when the hotword is detected)

Unfortunately, the Snowboy project failed for familiar reasons (lack of funding, thin development team, hard to scale, hard to sell), and AFAIK there aren't any FLOSS projects that have replaced it. What's even more unfortunate is that, if both Snowboy and Mycroft fail, the FLOSS community will have a hard time bridging the gap with commercial voice assistant products.

That's, of course, unless politics steps in, and they acknowledge that FLOSS solutions aren't competitive in a pure market-driven economy, and therefore if we want them to thrive and keep the lights on we need public money to support them (and a long-term strategy that decides which projects should receive the funding).

Double crumbs!
Making snowboys is probably the most difficult thing for me to do in this game.
#acnh #snowboy #AnimalCrossing #NintendoSwitch

Crumbs!
#acnh #snowboy

🔴 #Question J'ai un #RaspberryPi 2B et deux #OrangePi et j'aimerais bien avoir un home assistant façon #GoogleHome ou #Alexa.
Il y a eu plusieurs projets comme #Jarvis mais tous utilisaient #SnowBoy qui n'est plus disponible.

Avez-vous des pistes à me suggérer ?

Bien entendu, un truc avec commandes vocales en Français, si possible qu'on peut entraîner avec ses propres #HotWords !

Merci #Aide

:retoot: #RT #ReToot :retoot:

Modifying demo2.py to control a websocket. #snowboy #raspberrypi

Ok, so i can control with predefined sentences a #raspberrypi offline using #snowboy.

Time to control light!

First you need to record your sentence on https://snowboy.kitt.ai/dashboard (you can login using github).
Don't worry once you have download the pmdl files, you won't need internet anymore.

The more people repeat the sentence, the more models will be effective.

Ok, all seems fine, here is a scripts to install #snowboy on a #raspberrypi

https://gist.github.com/maditnerd/6ec136d282a646f21f11691522df8242

Stuck on the demo of #snowboy with a jack server error:

jack server is not running or cannot be started :control_knobs:

Demo for #snowboy didn't works. I created ~/.asoundrc and now it seems to works.

pcm.!default {
type asym
playback.pcm {
type plug
slave.pcm "hw:0,0"
}
capture.pcm {
type plug
slave.pcm "hw:1,0"
}
}

http://docs.kitt.ai/snowboy/#set-up-audio

Ok! time to download #snowboy

wget https://s3-us-west-2.amazonaws.com/snowboy/snowboy-releases/rpi-arm-raspbian-8.0-1.2.0.tar.bz2
tar -xvf rpi-arm-raspbian-8.0-1.2.0.tar.bz2

Tiens je testerais bien d'installer #snowboy sur un #raspberrypi en live sur #mastodon

I'm in the mood to try to install #snowboy on a #raspberrypi live on #mastodon.
https://mastodon.social/media/MT8opD4tN4cconG6dzU