Robust and Effective Noise Reduction of Speech Signal in Reverberant and Realistic Environment

Performance of current speech recognition systems severely degrades in the presence of noise and reverberation. While rather simple and effective noise reduction techniques have been extensively applied, coping with reverberation still remains as one of the toughest problems in speech recognition and signal processing. The objective of this paper is to provide improved real-time noise cancelling performance while keeping the high quality of enhanced speech by using new robust adaptive beam former. The beam forming approach is based on a fundamental theory of Normalized Least Mean Squares (NLMS) to improve Signal to Noise Ratio (SNR). The microphone has been implemented with a Voice Activity Detector (VAD) which uses time-delay estimation. To obtain a more robust feature against reverberation, the Linear Predictive (LP) residual calculation is performed.