Robust Feature Based on Speech Harmonic Structure for Speaker Identification
基于语音谐波结构的鲁棒特征参数及其在说话人识别中的应用

An effective and robust speech feature extraction method based on pitch frequency and harmonic structure is proposed by means of short-time spectrum analysis of clear and noisy speech. Experimental results indicate that the new feature is relatively insensitive to Additive White Gaussian Noise (AWGN). Compared to conventional cepstrums, the new feature can give outstanding improvement for closed-set text-independent speaker identification under noisy environments corrupted by AWGN.