6 ms·

For "actually serverless" voice chat, check out https://whisper.ggerganov.com/ https://whisper.ggerganov.com/

by lxe 4y ago

EGreg 4y ago
what is the point when the Web browsers mostly all already support it https://developer.mozilla.org/en-US/docs/Web/API/Web_Speech_API https://developer.mozilla.org/en-US/docs/Web/API/Web_Speech_...
- lxe 4y ago
  Engineering curiosity and fun
- hombre_fatal 4y ago
  When you want much better, state of the art quality. Compare it yourself.
  - EGreg 4y ago
    The browser seems to do very well for me, and in multiple languages. What about this, can it do languages?
    - RockRobotRock 4y ago
      Whisper supports many different languages.
    - hombre_fatal 4y ago
      Whisper can do you one better: you don't need to know the spoken language ahead of time, and you can mix languages in the same sentence. It does all sorts of things better like upgrading words into proper nouns, and doing it based on context. Honestly it's the first time speech recognition finally feels "production ready" after years of Siri/Alexa.
- easrng 4y ago
  In Chrome that's using Google's servers to transcribe the audio (last I checked) and it doesn't work on Firefox.
arthurcolle 4y ago
This is so good. I tried playing around with whisper.cpp last week and got absolutely terrible performance. I played around with the "step size" as well as "n" (not sure what these do, I should probably read the docs seriously). I tried a lot of tweaks but this one is definitely the best I've seen. How did it know how to spell my name correctly when I just spoke into my microphone??? The two L's usually trips up the transcription models. What????
- kurisufag 4y ago
  >How did it know how to spell my name correctly when I just spoke into my microphone??? if the variations are pronounced the same? luck, probably.
  - arthurcolle 4y ago
    My last name is pronounced "Collé" but when I've tried other whisper demos it always spelled it Cole or Coll. This is the first time I've ever seen it get the spelling actually right (missing the accent is forgivable) ;)
    - arthurcolle 3y ago
      After further testing, it seems like it is stochastic and I basically just got really lucky. Just wanted to post this for posterity. ¯\_(ツ)_/¯
- dom96 4y ago
  wow, you weren't kidding. I tried the tiny model and it got what I said perfectly. Super impressive.
- syntaxing 4y ago
  Hmm curious what the differences are, it’s the same person who wrote whisper cpp that hosts that website.
teacpde 4y ago
Really cool, I was able to play with it for a bit, but now seems like HN hug-of-death kicked in.
- lxe 4y ago
  It's hard to death-hug these wasm demos, as they are all static files that can be easily served via CDNs
antman 4y ago
Which of the models did you use?
- lxe 4y ago
  This isn't me, this is ggerganov's work. His demo is using gpt2, and whisper models (https://ggml.ggerganov.com/ https://ggml.ggerganov.com/).
jancsika 4y ago
This link is the most worthwhile hijack of an HN thread that I've ever seen. How is tiny.en so damned accurate? How much faster does this run natively? Do the NEON instructions work for arm devices like an RPI or is it just tuned for Apple? With this and Alpaca 13B you could probably replace an entire window manager. Edit: it seems like it was less than a year ago that local speech recognition was a slow slog through a swamp of crufty, complicated research projects with most the high quality training data hidden behind walled gardens. Now I can stutter-step over a word and a demo in my browser correctly transcribes what I meant sans stutter. What happened?
- lxe 4y ago
  SIMD support in wasm is a bit weird. You'll need to compile into multiple targets, and many compile flags are not supported: https://emscripten.org/docs/porting/simd.html https://emscripten.org/docs/porting/simd.html
- anotherhue 4y ago
  Yes NEON works on recent Pis. RPi CPU tuning: https://github.com/ggerganov/whisper.cpp/blob/70567eff232773d6786c91585d040f53c36b87a4/Makefile#L172 https://github.com/ggerganov/whisper.cpp/blob/70567eff232773...

For "actually serverless" voice chat, check out https://whisper.ggerganov.com/ https://whisper.ggerganov.com/

by lxe 4y ago

For "actually serverless" voice chat, check out https://whisper.ggerganov.com/ https://whisper.ggerganov.com/

what is the point when the Web browsers mostly all already support it https://developer.mozilla.org/en-US/docs/Web/API/Web_Speech_API https://developer.mozilla.org/en-US/docs/Web/API/Web_Speech_...

lxe 4y ago
Engineering curiosity and fun
hombre_fatal 4y ago
When you want much better, state of the art quality. Compare it yourself.
- EGreg 4y ago
  The browser seems to do very well for me, and in multiple languages. What about this, can it do languages?
  - RockRobotRock 4y ago
    Whisper supports many different languages.
  - hombre_fatal 4y ago
    Whisper can do you one better: you don't need to know the spoken language ahead of time, and you can mix languages in the same sentence. It does all sorts of things better like upgrading words into proper nouns, and doing it based on context. Honestly it's the first time speech recognition finally feels "production ready" after years of Siri/Alexa.
easrng 4y ago
In Chrome that's using Google's servers to transcribe the audio (last I checked) and it doesn't work on Firefox.

arthurcolle 4y ago

This is so good. I tried playing around with whisper.cpp last week and got absolutely terrible performance. I played around with the "step size" as well as "n" (not sure what these do, I should probably read the docs seriously). I tried a lot of tweaks but this one is definitely the best I've seen. How did it know how to spell my name correctly when I just spoke into my microphone??? The two L's usually trips up the transcription models. What????

kurisufag 4y ago
>How did it know how to spell my name correctly when I just spoke into my microphone??? if the variations are pronounced the same? luck, probably.
- arthurcolle 4y ago
  My last name is pronounced "Collé" but when I've tried other whisper demos it always spelled it Cole or Coll. This is the first time I've ever seen it get the spelling actually right (missing the accent is forgivable) ;)
  - arthurcolle 3y ago
    After further testing, it seems like it is stochastic and I basically just got really lucky. Just wanted to post this for posterity. ¯\_(ツ)_/¯
dom96 4y ago
wow, you weren't kidding. I tried the tiny model and it got what I said perfectly. Super impressive.
syntaxing 4y ago
Hmm curious what the differences are, it’s the same person who wrote whisper cpp that hosts that website.

teacpde 4y ago

Really cool, I was able to play with it for a bit, but now seems like HN hug-of-death kicked in.

lxe 4y ago
It's hard to death-hug these wasm demos, as they are all static files that can be easily served via CDNs

antman 4y ago

Which of the models did you use?

lxe 4y ago
This isn't me, this is ggerganov's work. His demo is using gpt2, and whisper models (https://ggml.ggerganov.com/ https://ggml.ggerganov.com/).

jancsika 4y ago

This link is the most worthwhile hijack of an HN thread that I've ever seen. How is tiny.en so damned accurate? How much faster does this run natively? Do the NEON instructions work for arm devices like an RPI or is it just tuned for Apple? With this and Alpaca 13B you could probably replace an entire window manager. Edit: it seems like it was less than a year ago that local speech recognition was a slow slog through a swamp of crufty, complicated research projects with most the high quality training data hidden behind walled gardens. Now I can stutter-step over a word and a demo in my browser correctly transcribes what I meant sans stutter. What happened?

lxe 4y ago
SIMD support in wasm is a bit weird. You'll need to compile into multiple targets, and many compile flags are not supported: https://emscripten.org/docs/porting/simd.html https://emscripten.org/docs/porting/simd.html
anotherhue 4y ago
Yes NEON works on recent Pis. RPi CPU tuning: https://github.com/ggerganov/whisper.cpp/blob/70567eff232773d6786c91585d040f53c36b87a4/Makefile#L172 https://github.com/ggerganov/whisper.cpp/blob/70567eff232773...