本文利用人耳的感知特性,提出了加权mel倒谱系数,并建立了相应的分析算法。实验结果表明,该系数不仅能够准确地刻画说话人声道的短时特征,还能用来重建出高质量的语音。因此加权mel倒谱分析不仅能够应用于语音识别和说话人识别,还能应用于语音编码和参数合成。
Based on auditory properties, the weighted mel-cepstral cocfficients (WMCC) are proposed along with the corresponding analysis method. Experiments show that not only the WMCC can represent the short-time characteristics of vocal of speakers, but also can be used for synthesizing the high-quality speech. Therefore the weighted mel-cepstral analysis can be applied to speech recognition, speaker identification, speech coding and parametric synthesis of speech.