Speech Recognition

Posted on
  • I just saw this: https://github.com/TalAter/annyang

    It's a 2kB JavaScript speech recognition library...

    Now I doubt it'll run on Espruino itself - but it doesn't mean you can't serve it up from Espruino, and then use your phone to control it with voice commands!

  • Good find. I could find use for this. "Lights on" or "Computer on" :-)

  • That's ace!

  • @DrAzzy's? - He had something like that going on quite a while ago to control the room lights... I saw it life in person.

  • I gave up. Never could get recognition rates that were vaguely acceptable. Chant the command for a minute, and have it register once.

    Fucking ripoff those boards were (easyvr 3.0).

  • It's a shame - EasyVR looked like a good idea (although seemed pretty heavyweight)

    Someone posted a while back about something that actually ran on Arduino - it appeared to be quite basic (looking for 'hard' and 'soft' sounds) but I got the impression it worked well if you chose the right words for it. I think it'd probably need running in native code on Espruino though.

    At least this one's a good halfway point - you get to use your phone, which has a mic and more than enough horsepower to do the calculation involved.

    edit: @DrAzzy you could take that Arduino-based software and make your own @AzzyVR boards with an AVR and a Mic - could be pretty popular :)

  • Dear DrAzzy,
    you are using quite offensive words.

    We are sorry that you had problems with your EasyVR 3 module and we would like to support you to solve them as our top priority is to have only happy customers :-)

    Our module does exactly what is intended for, nothing more and nothing less:

    Please contact us if you need support and we will do our best to help you.

    Thank you and kind regards,
    VeeaR Technical Support Team

  • I'm researching ways of having an offline ASR just recognise numbers, does anyone know of anything that could do that?

    Came across this https://www.matrix.one and https://snips.ai but it wasn't obvious to me if they could handle numbers. The Matrix sure looks a neat bit of tech though!

  • There are a few listed here as well: https://1sheeld.com/top-5-arduino-voice-­control-modules/

    Again, it's not obvious if they do numbers.

    This looks like a really interesting option too: https://github.com/arjo129/uSpeech (a library meant for Arduino)

    I'm not 100% sure if it could work on numbers though - it seems like it might actually only work on the very first sound in the word: https://github.com/arjo129/uSpeech/wiki/­Designing-your-vocabulary

  • Thanks for that @Gordon I spent yesterday getting my head around the Snips console and have trained an assistant that very accurately recognises numbers and various other keywords I've thrown in. I currently have it running on my Mac and am using JS to respond to intents. Would Snips run on any of your microprocessor devices?

  • Would Snips run on any of your microprocessor devices?

    Not directly as far as I know. If they actually output source code for their trained model then it may well be possible to compile it into Espruino though - it'd be a really neat trick.

    It might be worth asking them if it's possible?

  • ... and I just saw: https://petewarden.com/2019/03/07/launch­ing-tensorflow-lite-for-microcontrollers­/amp/

    Looks like they have training data which probably has numbers in, as well as pretty much everything you need to get some C++ files that you can build for a microcontroller: https://github.com/tensorflow/tensorflow­/tree/master/tensorflow/lite/experimenta­l/micro/examples/micro_speech

    I don't believe it would be particularly hard to get it built into Espruino.

  • I don't believe it would be particularly hard to get it built into Espruino.

    Maybe for you :) I don't know C++

  • Looks like there's a bit of interest in it, so when I get some time (maybe next week) I'll try and do a tutorial on it

  • Look forward to it.

  • Post a reply
    • Bold
    • Italics
    • Link
    • Image
    • List
    • Quote
    • code
    • Preview

Speech Recognition

Posted by Avatar for Gordon @Gordon