Carsten Sydow
Inst. for Electroacoust., Tech. Univ. of Darmstadt, Merckstr. 25, D-64283 Darmstadt, Germany
The signal-to-noise ratio of a speech signal picked up by a microphone array can be improved by adaptive post processing. Enhancement techniques known from single microphone or dual microphone signal processing, like noise canceling and spectral subtraction can be extended to a multimicrophone array system. The noise canceling technique and derived structures try to model the room impulse response by an adaptive transversal filter. Thus the performance of these algorithms is limited by the ratio of filter length to reverberation time and by the capability to track the nonstationary impulse response. Reduction of the noise of approximately 8 dB can be achieved with acceptable filter length in a stationary environment, but precautions must be taken to avoid canceling of the desired speech signal. The spectral subtraction method yields higher improvements in signal-to-noise ratio up to 12 dB, but due to false estimation of magnitude short time spectra the processed speech signal contains distortion known as ``musical tones.'' Different structures for adaptive post processing of microphone array signals based on both enhancement techniques are presented and discussed. Their performance in stationary and nonstationary environments is characterized by means of improvement of the signal-to-noise ratio and subjective speech quality.