语音分离与HMM相结合的语音增强方法

摘要

There are two typical speech enhancement algorithms based on HMM(Hidden Markov Model)which are MAP (Maximum A Posteriori)estimator and MMSE(Minimum Mean-Square Error)estimator. Both algorithms have high computa-tional complexity, and the former can’t handle non-stationary noise. In response to these shortcomings, with the speech separa-tion technology as reference, speech enhancement algorithm based on speech separation using HMM is designed. This algorithm uses the multi-state AR-HMM which is applied to non-stationary noise condition to decode the mixed state sequence of noisy speech under the speech model and noise model. Then, the decoded speech is estimated by speech separation method using maxi-mization model theory which avoids iterative procedure and huge computation so that the complexity is reduced. The experi-ments also show that the proposed algorithm can effectively remove the stationary noise and non-stationary noise, improve the PESQ(Perceptual Evaluation of Speech Quality)score and the algorithm time is under control too.%　　针对基于隐马尔科夫（HMM，Hidden Markov Model）的MAP和MMSE两种语音增强算法计算量大且前者不能处理非平稳噪声的问题，借鉴语音分离方法，提出了一种语音分离与HMM相结合的语音增强算法。该算法采用适合处理非平稳噪声的多状态多混合单元HMM，对带噪语音在语音模型和噪声模型下的混合状态进行解码，结合语音分离方法中的最大模型理论进行语音估计，避免了迭代过程和计算量特别大的公式计算，减少了计算复杂度。实验表明，该算法能够有效地去除平稳噪声和非平稳噪声，且感知评价指标PESQ 的得分有明显提高，算法时间也得到有效控制。