An Improved Design for Speech Denoising

Abstract

In high-noise environment, the background noises have greatly bad influence on the speech recognition and its transmission quality. The speech extraction and the noise reduction are long-term questions which the communication system has not solved effectively all the while. In this paper, we propose an improved system solution based on the combination of the bone-conduction sensor, a digital signal processor contained denoising algorithms and an active noise reduction part. At the beginning, the bone-conduction sensor is used to gather the voice. Next, the collected noisy speech signal is enhanced with spectral subtraction algorithm. Then the enhanced signal is detected with the improved short-time energy voice activity detection algorithm based on the Mel frequency. Finally we employ the ANR technology in the receiver. We have designed a prototype system based on the wireless platform. The results of tests show that it achieves better progress in anti-noise performance.