Rank-order speech classification algorithm (RASCAL)

This paper describes a theoretical and experimental investigation for detecting the presence of speech in wideband noise. A robust algorithm for making the silence-voiced-unvoiced decision is described. This algorithm is based on a nonparametric rank-order statistical signal-detection scheme that does not require a training set of data and maintains a constant false-alarm rate for a broad class of noise inputs corresponding to a single decision threshold. The nonparametric rank-order decision procedure is the multiple use of the two-sample Savage T statistic. The performance of this detector is evaluated and compared to that obtained by manually classifying twenty recorded utterances with 39, 30, 20, 10, and 0 decibel signal-to-noise ratios. In limited testing, the average probability of misclassification is less than 5 percent, 12 percent, and 55 percent for signal-to-noise ratios of 39, 20, and 0 decibels respectively.