Study and Application of Silence Model Adaptation for Use in Telephone Speech Recognition System

This paper addresses the problem of the mismatch between a silence model and background noises which often occurs in a telephone speech recognition system (SRS) application. At first, the use of parallel model combination (PMC) methods is studied with the respect to this application. Secondly, the effective adaptation of a silence model to various background noises is confirmed. Finally, an original method combining log-add PMC with a noise power spectral density estimation based on minimum statistics is proposed. The performed tests prove the benefit of the suggested method to the speech recognition results that is caused by the stability of speech vector selection under the influence of various background noises. The advantages can be seen in no extra voice activity detector and in a relatively low computational load.