Shop OBEX P1 Docs P2 Docs Learn Events
vocal tract help! — Parallax Forums

vocal tract help!

SovrSovr Posts: 12
edited 2011-06-12 01:04 in Propeller 1
hello everyone :-), I need some help on how to use the vocal tract propeller code sample. I have been doing spin code for almost three years now, but I have been wanting to make a robot that understands what I say. If anyone can show me a code that shows me how to work with it that would be great! (to get me started). Thank you very much!:smile:

Comments

  • Mike GreenMike Green Posts: 23,101
    edited 2011-06-11 20:49
    The vocal tract code has nothing to do with a robot that understands what you say. It synthesizes speech from vocal tract parameters and, with Phil Pilgrim's program, can use phonemes as the input. There was some completely unrelated code posted in this forum attempting to recognize a few spoken words. I don't have a link to that, perhaps someone else does.
  • Jack BuffingtonJack Buffington Posts: 115
    edited 2011-06-11 23:14
    I was hoping that this would be a question that I could actually answer but alas... Right now I am deep into the vocal tract trying to create a better text to speech program. So far so good but I have a lot of work ahead of me. I will say that if you set out to write your own program to generate speech sounds, you will have a pretty good idea of what you need to do to recognize those sounds too. *Maybe* some day I'll take a stab at that. It would be pretty cool to have speech recognition on a propeller (realistically probably a propeller II). Anyway, read up on formants, Fourier transforms and its variants. With a good understanding of those, you could probably recognize most or all of the phonemes. It would be a good idea to pay much more attention to frequencies below about 2500 Hz as that is where all of the formants seem to lie. The fricatives tend to spread out over large areas of the spectrum so you could probably just test a few points above 2500 Hz to figure out which one was being pronounced at the time.

    Someone was doing Fourier transforms on the propeller but at a pretty slow update rate as I recall. Maybe it would be sufficient to catch everything but sounds like K and P? If not then you'll just have to wait until the propeller II comes out or give a look at Leon's favorite propeller alternative.

    I have been using Audacity to do a lot of my work with the vocal tract. With it, you can get a spectrograph of your voice to see what is going on when you talk. It is helpful but it would be nice if they had more resolution in the lower frequencies.
  • LeonLeon Posts: 7,620
    edited 2011-06-12 00:15
    A speech recognition library is available for the Microchip dsPIC devices:

    http://www.microchip.com/stellent/idcplg?IdcService=SS_GET_PAGE&nodeId=1406&dDocName=en023596

    It uses the usual Hidden Markov technique.
  • Phil Pilgrim (PhiPi)Phil Pilgrim (PhiPi) Posts: 23,514
    edited 2011-06-12 01:04
    Here's my so-so stab at speech recognition:

    -Phil
Sign In or Register to comment.