摘要

The purpose of this research is to develop an algorithm which can be used in the system of speech recognition. In the previous research, a target signal can be reconstructed by a signal separation technique which is based on time-reversal method (TRM) via self-focusing property. In this research, a multiple signal separation technique which is based on adaptive time-reversal method (ATRM) is developed. The advantage of ATRM is to reconstruct target signal and to cancel noise signals simultaneously if impulse response functions of environment between target source, noise sources, and array microphone are known. In the simulation, correlation coefficient and signal to noise ratio (SNR) are implemented as indicators to evaluate the performance of signal separation for TRM and ATRM in a reverberation room, and to investigate the differences between two methods. The comparison indicates that number of array microphone is significantly reduced from 14-microphones via TRM to 6-microphones via ATRM. Furthermore, the experiments of signal separation are conducted to a model of smart TV with multiple noise sources in the environment. Finally, a real time system of signal separation was developed. All the results indicate that the target of source was clearly reconstructed with enhancing of correlation coefficient and SNR by using ATRM.

  • 出版日期2014-10-1