A speech enhancement algorithm based on a non-negative hidden Markov model and Kullback-Leibler divergence

Abstract In this paper, we propose a supervised single-channel speech enhancement method that combines Kullback-Leibler (KL) divergence-based non-negative matrix factorization (NMF) and hidden Markov model (NMF-HMM). With the integration of HMM, temporal dynamics information signals can be taken into account. This includes training stage an stage. stage, sum Poisson distribution, leading to KL divergence measure, is used as observation for each state HMM. ensures computationally efficient multiplicative update parameter model. online novel minimum mean square error estimator proposed NMF-HMM. implemented using parallel computing, reducing time complexity. Moreover, compared traditional NMF-based methods, experimental results show our algorithm improved short-time objective intelligibility perceptual evaluation quality by 5% 0.18, respectively.