Spectral subtraction using reduced delay convolution and adaptive averaging

In hands-free speech communication, the signal-to-noise ratio (SNR) is ofte n poor, which makes it difficult to have a relaxed conversation. By using n oise suppression, the conversation quality can be improved. This paper desc ribes a noise suppression algorithm based on spectral subtraction. The meth od employs a noise and speech-dependent gain function for each frequency co mponent. Proper measures have been taken to obtain a corresponding causal f ilter and also to ensure that the circular convolution originating from fas t Fourier transform (FFT) filtering yields a truly linear filtering. A nove l method that uses spectrum-dependent adaptive averaging to decrease the va riance of the gain function is also presented. The results show a 10-dB bac kground noise reduction for all input SNR situations tested in the range -6 to 16 dB, as well as improvement in speech quality and reduction of noise artifacts as compared with conventional spectral subtraction methods.