基于HFM和感知后滤波器的语音增强

研究了只能获得带噪信号的情况下的语音增强问题。将语音信号看作由高斯噪声激励的自回归(AR)过程,观测噪声为加性高斯白噪声,把信号转化为状态空间模型。首先用隐马尔可夫模型(HMM)估计AR参数和噪声的方差作为卡尔曼滤波器初值,估计信号作为参数估计的中间值给出,然后将估计信号通过一个感知滤波器平滑以消除残余噪声。仿真结果表明该算法有良好的性能。

Speech Enhancement Based on HFM with Perceptual Post-Filtering

This paper discusses speech enhancement when only noisy speech signal are available.The speech signal is modeled as an autoregressive(AR)process excited by a Gaussian signal.And the observation noise is an additive white Gaussian noise.Then the model is represented in state space form.AR parameters and noise covariance as the initial of Kalman filter are estimated using Hidden Markov Model(HMM).The sig- nal is a byproduct of the parameter estimation process.And then the estimated signal is smoothed by a perceptu- al filter to eliminate noise residual.Simulation results show the method has good performance.