论文部分内容阅读
语音识别系统的识别率在噪声环境中下降很大。本文根据人耳的听觉特性,提出一种基于人耳听觉掩蔽特性的抗噪声特征提取方法。该方法先求取噪声语音的掩蔽特性,在此基础上再计算Mel倒谱系数用于语音识别。通过对TIMIT数据包的 0~9十个英语数字在 NoiseX92的各种噪声下进行了识别试验。其中在信噪比 0dB条件下,在 3种噪声条件下识别率平均提高 152%,实验表明新方法对于各种噪声环境下的识别率有显著提高。
Recognition rate of voice recognition system in the noise environment decreased significantly. In this paper, based on the auditory characteristics of the human ear, an anti-noise feature extraction method based on human auditory masking characteristics is proposed. The method firstly obtains the masking property of noise speech, and then calculates the Mel cepstrum coefficient for speech recognition. Through the TIMIT packet of 0 to 9 ten English digital noiseX92 in a variety of noise under the recognition test. The recognition rate under the three kinds of noise conditions is improved by 152% on the condition of 0dB signal-to-noise ratio. Experiments show that the new method can improve the recognition rate under various noise environments.