VCP200 Speaker-Independent  (276-1308)    Features     Faxback Doc. # 15361
Word Recognizer

The VCP200 is a speech recognition IC that understands a limited number of
spoken commands, regardless of the person speaking.  It identifies commands
by breaking the words into broad phoneme classes, and then recognizing
predetermined strings of these classes.  The VCP200 performs a spectral
analysis of the incoming speech signal over the range of 300 Hz to 5500
Hz, determines the membership of phoneme classes based on spectral shape,
forms strings of these classes, and compares the strings to the stored
strings of selected commands.  The VCP200 recognizes the selected commands,
regardless of who speaks them, in real time.

The VCP200 has two speech recognition modes.  In the Command Mode, it
recognizes five motion commands:  go, stop, left turn, turn right, and
reverse.  In the Yes-No/On-Off Mode, it recognizes two commands:  either
yes and no or on and off.  When the given command is recognized, the
associated output for that command provides a latched logic low, capable of
sinking up to 10mA.  In the Yes-No/On-Off Mode, a third output indicates no
recognition.

The VCP200 is designed for use in a relatively quiet environment.  Constant
background noise such as caused by fans, air conditioning, or motors can
usually be accommodated by adjusting the input amplifier's gain so that the
output signal at the second amplifier does not exceed 1 volt peak-to-peak
with no speech input.  Intermittent noises such as slamming doors, loud
music, or construction sounds are harder to deal with .  A close-talking,
noise-canceling microphone helps, although occasional misrecognition might
still occur.  In general, the lower the level of outside noises, the better
the VCP200 performs.

The location of the microphone relative to the person speaking also affects
performance and is dependent on the background noise.  The quieter the
noise level, the greater the microphone's usable distance.  Good
performance at up to 2 feet is common for quiet areas.  It might help to
experiment with several microphones to optimize performance.  It might
also help to change the frequency/gain characteristics of the input
amplifiers to compensate for the high-frequency loss (greater than 4 kHz)
as distance increases.

While the VCP200 ignores some words not in its vocabulary, it was designed
with the assumption that it need only discriminate among the programmed
commands.  As a result, words that are not in the VCP200's vocabulary are
sometimes identified as being commands, but commands are rarely confused
with each other.  If the VCP200 must function in the presence of other
spoken words, it should be disabled until a commands is expected by either
bringing Pin 20 low or by using a push-to-talk switch on the microphone.

Its features include:

    Recognizes five motion commands:  go, stop, reverse, turn right, and
    left turn or two commands:  yes and no or on and off.

    Active low outputs can sink up to 10mA

    Powered from a single 5V supply.

    Uses clipped analog input so you do not need an analog-to-digital
    conversion circuit.

    Requires minimal support hardware.


(ALL-05/25/95)