Shop OBEX P1 Docs P2 Docs Learn Events
Speech recognition (Has it ben done?) - Page 4 — Parallax Forums

Speech recognition (Has it ben done?)

124»

Comments

  • HumanoidoHumanoido Posts: 5,770
    edited 2008-07-09 00:39
    Using the Basic Stamp 2, you could look at
    capturing .5 second of sound frequency, using
    a mic and frequency chip, then buffering the result
    through some software conditioning to
    perhaps recognize a burst of sound and simple
    short one word commands.

    humanoido
  • Mike GreenMike Green Posts: 23,101
    edited 2008-07-09 00:53
    The BS2 really isn't a good platform for this because of the very limited data storage available. Even a 1/2 second of sound takes a lot of storage. The Apple II (and other similar early home computers) used to do this with a 1K (???) buffer for the raw sound samples, then would compress this to a pattern. I forget how large that was, but it was more than the 26 bytes available with a Stamp.
  • HumanoidoHumanoido Posts: 5,770
    edited 2008-07-09 02:50
    Agreed, it would be challenging. I remember building one for the TS-1000 computer but it didn't need 1K of RAM to operate. The BS2 can frequency sample in real time within the frame of 1/2 second, average and data log into the external chip providing that timing constraints can be met. The data can be read back and the sound parameters reconstructed. This would encompass a compression ratio. There's 128 bytes RAM on the BS2px and the clock is much faster. The signal could be sampled "during" the half second frame, and spooled as data. The data would need reconstruction to reproduce the sound. Speech can handle low quality sample rates. A 4-bit encoder is capable of generating a minimal FR and software can enhance the signal with conditioning. The same stamp also has a serin of 19,200 baud and some possibilities to explore with minimal external components. With some time, I'll take a look at creating one using a Penguin Robot and AUX5.

    humanoido
Sign In or Register to comment.