论文部分内容阅读
提出了一种复杂场景下基于MFCC与基频特征贡献度的说话人性别识别方法.该方法有效融合了基于Mel频率倒谱系数的模板匹配方法和基音频率判别方法.实验语音数据库包括5 000个孤立词语音和1 260个带情感的语音.在安静环境下说话人的性别识别率可以达到98.88%,在信噪比为10dB的babble噪声下通过谱减法降噪后的识别率为90.2%.实验表明:说话人情绪对性别识别的影响较大,尤其是男声.
A speaker gender recognition method based on the contribution of MFCC and fundamental frequency features in complex scenes is proposed.This method effectively combines template matching method and pitch frequency discrimination method based on Mel frequency cepstral coefficient.The experimental speech database includes 5 000 Isolated speech and 1 260 emotive speech.Under a quiet environment, the speaker’s gender recognition rate can reach 98.88%, and the recognition rate after spectral subtraction noise reduction under a babble noise with a signal-to-noise ratio of 10dB is 90.2%. Experiments show that the speaker’s mood has great influence on gender identification, especially male voice.