Shop OBEX P1 Docs P2 Docs Learn Events
Voice samples needed, any volunteers? — Parallax Forums

Voice samples needed, any volunteers?

lonesocklonesock Posts: 917
edited 2009-10-27 19:53 in Propeller 1
Hi, Everybody.

I am looking to do some simple voice recognition on the propeller (see this thread: http://forums.parallax.com/showthread.php?p=848739. If anyone has any questions please ask in the other thread [noparse][[/noparse]8^). I would like to start collecting a mini-corpus of voice samples, and I need some volunteers. If you are willing, here is what I would ask:

* Please record yourself in a quiet environment
* please use a sample rate of 16kHz or above
* for each of the following 6 words, "up,down,left,right,yes,no", please
-- say each word 5 times, with ~2 second pause between each word
-- save that recording as whatever_word.wav
-- upload the 6 wav files in a response post, preferably all zipped
* feel free to use your name as the zip filename if you want to be credited in the final source

The resulting library (assuming it works) will be uploaded to the OBEX under the MIT license.

Thank you all in advance!
Jonathan

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
lonesock
Piranha are people too.

Comments

  • SRLMSRLM Posts: 5,045
    edited 2009-10-24 06:56
    Hi lonesock,

    I did some samples. I don't know if the sample rate is high enough; I don't have any programs or anything to tell me that. Let me know if it doesn't work out, and I can try a different setup.

    SRLM
    zip
    zip
    SRLM_voice_samples.zip
    592K
  • lonesocklonesock Posts: 917
    edited 2009-10-25 01:32
    SRLM said...
    Hi lonesock,

    I did some samples. I don't know if the sample rate is high enough; I don't have any programs or anything to tell me that. Let me know if it doesn't work out, and I can try a different setup.

    SRLM
    Hi, SRLM.

    Thanks, the sample rate is more than sufficient (48kHz)!

    If anyone would rather email me the zip instead of post it here, that's fine too. My email is this user name at gmail.com.

    thanks!
    Jonathan

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    lonesock
    Piranha are people too.
  • DelusDelus Posts: 79
    edited 2009-10-25 03:18
    Hello Jonathan

    I would love to contribute to this project, might even get a few friends to pitch in. I'm not sure how you plan to make a generalized algorithm for this but even with just 6 commands this would be a great addition to the prop. I'll post my own sample tomorrow and ask a few others if they'd be interested.

    David
  • lonesocklonesock Posts: 917
    edited 2009-10-25 07:37
    Thanks, David, both for your samples and your recruiting...the more the merrier! The link in the 1st post explains my basic plan regarding the algorithms in use, etc. Obviously no guarantees yet, but hopefully it will work [noparse][[/noparse]8^). I'm building up the prop code (to see my processing limitations on the actual hardware) and some C++ code (for ease of testing, debugging, etc.) simultaneously. The next step is running the data reduction algorithms (already in place) on a bunch of actual voice samples from different speakers (hence this post), to get a feel for which reduction routines help identify each word in a speaker independent manner, and how important each is. It is quite possible that some changes will be made to the current code.

    thanks,
    Jonathan

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    lonesock
    Piranha are people too.
  • DelusDelus Posts: 79
    edited 2009-10-26 14:44
    Ok ... I'm a day behind but here's my sample and one of a friend who agreed to help.

    David

    Post Edited (Delus) : 10/27/2009 12:50:44 AM GMT
  • DelusDelus Posts: 79
    edited 2009-10-26 14:57
    apologies it seems my recording program did something strange when it saved those files. I'm out of time just now but I'll see if I can redo them tonight.
  • Toby SeckshundToby Seckshund Posts: 2,027
    edited 2009-10-26 21:27
    Do you want to test things with an English, west country accent ?

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    Style and grace : Nil point
  • lonesocklonesock Posts: 917
    edited 2009-10-26 21:39
    @David: Not a problem, thanks for your help!
    Toby Seckshund said...
    Do you want to test things with an English, west country accent ?
    Please! Actually, we should do one better...the _real_ test of voice recognition is to recognize, then finish, any quote from Monty Python and the Holy Grail. (An acceptable substitute would be the original Hitchhiker's Guide radio drama.)

    Jonathan

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    lonesock
    Piranha are people too.
  • Toby SeckshundToby Seckshund Posts: 2,027
    edited 2009-10-27 09:32
    About 6 years ago, my boss and me both thought about legging it out to Spain. It started a whole series of spanish text messages, back and forth, usually along the lines of "My hovercraft is ..."

    I'll see if I can get to a quiet location for recording (there always seems to be too much noise at home, and I'm half deaf, Hey-ho)


    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    Style and grace : Nil point
  • lonesocklonesock Posts: 917
    edited 2009-10-27 16:54
    Toby Seckshund said...
    About 6 years ago, my boss and me both thought about legging it out to Spain. It started a whole series of spanish text messages, back and forth, usually along the lines of "My hovercraft is ..."
    [noparse][[/noparse]8^) So, informal poll, should we add "eels" to the list of recognized words?

    Jonathan

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    lonesock
    Piranha are people too.
  • Toby SeckshundToby Seckshund Posts: 2,027
    edited 2009-10-27 19:53
    Just brought home a mic ad put it through the mixer and ...

    The Bunnies, in the same room (don't ask) started to crash, bang and thump. It a conspirisy !!!!

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    Style and grace : Nil point
Sign In or Register to comment.