Evaluation of Speech Enhancement Techniques for Speaker Identification in Noisy Environments

In automatic speaker recognition applications, the presence of background noise severely degrades the performance of such systems. One solution to this problem is to use speech enhancement techniques aimed at reducing the acoustical noise in the speech signal, applied prior to the speaker recognizer. In this paper, we evaluate the impact of different speech enhancement techniques for robust speaker identification. We use clean speech corpus from TIMIT database and combine the speech signal with different types of noise from the NOISEX-92 database. Our results show that better speaker identification rates are attainable under mismatched conditions especially at low signal-to-noise ratios (SNRs).