A simple adaptive filtering technique for speech enhancement

A simple technique for speech enhancement has been developed which does not require that speech and noise inputs be available separately. An adaptive predictor DPCM algorithm is used to derive the reference input to an adaptive filter. The DPCM algorithm generates a signal which is correlated with the original speech waveform but uncorrelated with the noise. The system has been shown to enhance perceptual speech quality, particularly at low input SNRs.