A variance detector for signal-gap discrimination in noisy speech channels
James L. Wayman · Defense Technical Information Center (DTIC) · 1985
This report documents a program written for post- (FFT) processor activity/gap detection in noisy speech communication channels. The algorithm used is based on the assumption that the variance of the logs of the frequency-domain magnitudes of a signal is greater than the corresponding variance of the channel noise. Trade-offs between Type I and Type II errors can be controlled with and interactively defined program parameters. Parameters were found to yield a 16% missed detection rate with a 5% false alarm rate, performance entirely acceptable in speech compression applications. Some problems were encountered with sibilate detection, due to the close relationship between these sounds and white noise. (Author)