6 votes

An All-Neural On-Device Speech Recognizer

3 comments

  1. Wes Link
    That's pretty remarkable. I'm glad to be reducing reliance on a network for what is becoming a much more frequent and natural method of input. I hope that high-quality offline text-to-speech...

    That's pretty remarkable. I'm glad to be reducing reliance on a network for what is becoming a much more frequent and natural method of input.

    I hope that high-quality offline text-to-speech follows. PicoTTS works, but it's not great. Getting the quality of wavenet voices onto local devices would be another leap forward.

    4 votes
  2. chas Link
    Good. Cloud-based speech recognition is an abomination, in terms of privacy, reliability, and - in certain regards - performance.

    Good. Cloud-based speech recognition is an abomination, in terms of privacy, reliability, and - in certain regards - performance.

    2 votes
  3. Neverland (edited ) Link
    Just yesterday I was thinking that I would love a google home or Alexa style device in my house, but I don't want the recordings leaving my house. So this seems like a great step in the right...

    Just yesterday I was thinking that I would love a google home or Alexa style device in my house, but I don't want the recordings leaving my house. So this seems like a great step in the right direction, right?

    I realize that I'd have to pay more for a device like that, but it would be worth it.

    edit: ooo, I found this project which claims to be all about privacy-first. https://snips.ai/ Does anyone know anything this and how well it works?

    Edit2: this project looks like actual FOSS https://github.com/leon-ai/leon

    2 votes