Method for estimating noise power spectrum and voice activity

Abstract

The invention relates to a method for estimating the noise power spectrum and the voice activity. According to the method, the appear probability of a voice on a frequency sub band and power spectrum information of noise can be finally deduced according to the time sequence relevance of a sequential hidden markov model (SHMM) description language based on first-order regression on each frequency component. The method comprises the following steps of: 1) extracting a logarithmic amplitude spectrum envelop for a voice signal on each frequency component, and constructing a corresponding binary hidden markov model, wherein each state is represented by Gaussian distribution; 2) for a field of voice data, setting M frames of caches, storing the previous M frames of input signals into the caches, extracting M frames of logarithmic amplitude spectrums from the caches, and constructing an initialization model by adopting a maximum likelihood estimation algorithm; and 3) after the initialization model lambdaM is obtained, starting from the (M+1)th frame, gradually updating the HMM of each frequency band by adopting an incremental learning method, and sequentially performing recurrence to obtain a noise value and the appear probability of a voice signal.

Description

technical field [0001] The present invention relates to the technical field of speech signal processing, in particular, the present invention relates to a noise spectrum estimation and speech activity detection method based on a sequential hidden Markov model. Among them, voice activity detection is an algorithm for judging the presence or absence of voice in the time dimension. It can not only answer the presence of voice in the form of "yes" or "no", but also describe the presence of voice with the probability of voice occurrence. Background technique [0002] Voice activity detection and noise power spectrum estimation are essential components of noise reduction algorithms, and their performance directly affects the performance of noise reduction algorithms, especially in harsh noise environments, where they indirectly affect speech processing systems (such as speech recognition, speaker recognition, and speech recognizer) performance. [0003] Most speech application sy...

AI Technical Summary

Problems solved by technology

This method has not been mentioned in previous literature

[0005] 2. The parameter adaptation of the traditional sequential HMM adopts a high-order regression average method. The current HMM parameter set depends on the model at the previous moment, the current observation value, and the observation values ​​at multiple moments in the past. This parameter regression The calculation method is huge

Its defect is: in some applications, this assumption is difficult to be satisfied, for example, when a sentence begins with a speech signal, it will cause the initialization of the noise model to fail, which in turn makes speech detection and noise power spectrum estimation inaccurate

Images