SPEECH ENHANCEMENT UNDER REVERBERANT CONDITION USING LINEAR PREDICTION. RESIDUAL SIGNAL

Abstract

Generally, reverberant speech is processed by passing it through inverse filter of room response. But the accurate realization of the inverse filter of acoustic impulse of a room is almost impossible. Further, any movement between source and microphone is the cause of change in the room impulse response drastically. So the performance of the inverse filtering is not good enough when impulse response is unknown. Also in inversing filtering, prior knowledge of room impulse response is necessary. So to avoid these difficulties here a different approach for speech processing degraded by reverberation is used. This method is based on analysis of short (2ms) segments of speech data to enhance the regions in the speech signal having high signal-to-reverberant component ratio (SRR). The short segments analysis shows that signal-to-reverberant component ratio is different in different segments of speech. The processing method involves identifying and manipulating the liner prediction residual signal in three different regions of speech signal, namely, high SRR region, low SRR region, and only reverberation component region. A weight function is derived to modify the linear prediction residual signal. The weighted residual signal samples are used to excite a time-varying all-pole filter to obtain perceptually enhanced speech signal. The performance of this method is illustrated through subjective and objective evaluations.