DPRG List  

DPRG: voice recognition

Subject: DPRG: voice recognition
From: Matt Minnis mminnis at prefres.com
Date: Tue Feb 2 08:07:51 CST 1999

A method that I saw somwhere, I think in a Circuit Cellar Book, did this a
little diffrent.

What he did was have the robot ask questions and then he would answer yes
or no.
He got very good results with this method.

Perhaps with this more modern circuit, this could be expanded upon.

Just some thoughts.


Matt Minnis

>The chip itself had good reproducability if you said the word during
>recognition mode exactly like you said it in training mode because it
>seemed to pick up on inflections and tone of the trainer.  I always kept
>myself close to the microphone because I didn't amplify the signal going
>into the chip (I wanted to just see if it would work initially).  Now
>that it works, i am integrating other behaviors before I start messing
>with the details of noise reduction and amplification.  I don't think
>the chip itself is very good with background noise, but I believe that
>the problem could be fixed on the microphone end before the signal goes
>into the chip.  The reliability was surprisingly good.  The circuit
>itself uses RAM so when you turn the power off, you have to retrain, but
>that is no big deal because you can hook both the processor and the
>keypad to the chip at the same time, and if you have a switch
>determining the mode of the chip, everything is trivial.  I had a switch
>allowing power to the RAM and not to the chip and auxilliary chips so
>that I wouldn't have to retrain while being able to power down the rest
>of the circuit.  Yeah, there were a few errors in recognition when the
>words were similar or I said the word in a completely different way, but
>all in all, the circuit was amazingly successful.  I know I was

Preferred Resources          (314) 567-7600 phone
701 Emerson rd.              (314) 993-6699 fax
Suite 475
St. Louis, MO


More information about the DPRG mailing list