论文部分内容阅读
为了提高基于隐含Markov模型的语音识别系统在噪声环境中的稳健性,研究了一种融合语音增强与后续补偿的抗噪声语音识别方法。在前端,语音增强有效地抑制背景噪声,从而提高了输入信号的信噪比。语音增强后的剩余噪声以及语音失真是对语音识别不利的因素,其影响将通过识别阶段的并行模型合并或特征提取阶段的倒谱均值归一化得到补偿。实验结果表明,此方法能够显著地提高语音识别系统在噪声环境中,特别是低信噪比条件下的识别精度,如对-5dB的白噪声,该方法可将识别精度从11.7%提高至71.0%。
In order to improve the robustness of speech recognition system based on implicit Markov model in noisy environments, an anti-noise speech recognition method based on implicit speech enhancement and subsequent compensation is proposed. At the front end, speech enhancement effectively suppresses background noise, thereby increasing the signal-to-noise ratio of the input signal. Speech-enhanced residual noise and speech distortion are unfavorable factors for speech recognition, and their effects are compensated for by normalizing the cepstral mean of parallel model incorporation or feature extraction phases in the recognition phase. Experimental results show that this method can significantly improve the recognition accuracy of speech recognition system in noisy environments, especially in low signal-to-noise ratio, such as -5dB white noise. The proposed method can improve the recognition accuracy from 11.7% to 71.0 %.