Shop OBEX P1 Docs P2 Docs Learn Events
Speech Input — Parallax Forums

Speech Input

HumanoidoHumanoido Posts: 5,770
edited 2011-02-12 22:45 in Propeller 1
I realize Parallax has a product to do this, but consider a different approach.
I'm curious about the Parallax Demo Board which has a microphone.
Is it possible to create a speech recognition program for only a few words?

I remember the 2K version of the TS-1000 computer back in the 1980s. We
could attach a microphone through the cassette recorder jack (same as a mic
input) and digitize some words. The number of words was not a lot, i.e. I only
had to use 4 words for a robot.

Thinking that the Propeller is so fast and has 32K HUB RAM, wouldn't it
also have the capability to recognize a few words in a very minimal
system?

Comments

  • Mike GreenMike Green Posts: 23,101
    edited 2011-02-12 08:32
    Yes. The Prop could do very simple speech recognition. The Apple II had simple speech recognition using a 6502 microprocessor and not a lot of memory. It recorded 8 bit sound in a buffer, normalized it, compressed it, and compared the result to a dictionary of previously recorded compressed sound samples. The accuracy was terrible although a very few carefully chosen words worked well enough for impressive demonstrations.
  • Phil Pilgrim (PhiPi)Phil Pilgrim (PhiPi) Posts: 23,514
    edited 2011-02-12 09:05
    There's a little bit here: http://forums.parallax.com/showthread.php?115725

    -Phil
  • BotdocterBotdocter Posts: 271
    edited 2011-02-12 11:54
    There is a complete working speech recognition engine in the obex.

    http://obex.parallax.com/objects/565/
  • HumanoidoHumanoido Posts: 5,770
    edited 2011-02-12 22:45
    Botdoctor: Thanks for this great find by Microcontrolled. It works on the demo board and samples vs memory are adjustable. It's a program that could be tuned to achieve a greater recognition rate.

    Mike Green: Thanks for this positive reply and the bit of info about the Apple II. I remember speech "output" on Apple II using S.A.M. Software Automated Mouth. It was one of the coolest programs I ever experienced. Now we can do that kind of speech on the Propeller. TRS-80 had a speech recognition unit that was very accurate (you could triple train fewer words for more accuracy) but it was a separate box. I really like the clever software only approach that uses resources of the computer.

    Leon: Thanks for finding another speech input method. This is a good option. However, everything from Dr. Jim comes with a price and is not open. I prefer to stick with open source because my projects are open source.

    Phil: What can I say? Your code is exactly what I had in mind. This program demonstrates great capability and resources of the Propeller chip, and talent + skill that lurks inside your vast programming mind! Thank you for making this available open source.
Sign In or Register to comment.