| 摘要 |
<p>In a speech recognition system for programming or commanding computers and other machines, there is described a recognition system which is less susceptible to breathing and other background noises. Speech-like signal durations are detected from input signal waves from all sources. The waveforms of the input signals are analyzed and recognition parameters are extracted. A spectral change detecting unit detects the magnitude of shorttime spectral changes from the speech-like signal durations, determining thereby whether the speech-like signal is speech. A recognition unit recognizes speech patterns from signals within the speech-like signal durations and rejects those signals which the spectral-change detecting unit indicates do not contain speech. The parameter extracting means and the spectral change detecting unit receive controlling signals from a control unit.</p> |
