论文部分内容阅读
对于基于连续隐马尔可夫模型（CHMM）的语音识别系统，为了提高系统在环境噪声下的鲁棒性，本文提出了一种能有效抑制加性平稳噪声和通道卷积噪声的相对自相关序列的Mel倒谱参数（RAS_MFCC＋△RAS_NFCC），进行特征参数级的去噪，明显地改善了系统的噪声鲁棒性。为了进一步提高系统在低信噪比语音时的识别性能，我们采用了CHMM的混合语青训练法，获得了对各种信噪比语音都具有很强适应性的CHMM参数。实验证明，这种将特征参数级去噪和系统模型级补偿相结合的方法明显地提高了语音识别系统的识别性能和抗噪性能
For the speech recognition system based on continuous hidden Markov model (CHMM), in order to improve the robustness of the system under ambient noise, a relative autocorrelation sequence that can effectively suppress additive stationary noise and channel convolution noise Mel-cepstral parameters (RAS_MFCC + △ RAS_NFCC) are used to denoise the feature parameters level, which obviously improves the noise robustness of the system. In order to further improve the performance of the system in recognizing speech with low signal-to-noise ratio, we adopt the hybrid language youth training method of CHMM to obtain the CHMM parameters which have strong adaptability to various SNR signals. Experiments show that the method of combining feature parameter level denoising and system model level compensation significantly improves the recognition performance and noise immunity of the speech recognition system