Speech enhancement method in music background based on non-negative matrix factorization

Abstract

The invention discloses a speech enhancement method in a music background based on non-negative matrix factorization, and belongs to the field of voice analysis or synthetization, audio analysis or processing. The method provided by the invention comprises the steps of framing and windowing a mixed signal of music and voice, carrying out non-negative matrix factorization to an STFT (The Short-Time Fourier Transform) amplitude spectrum, wherein a basic matrix of the background music is obtained by training and is fixed in a decomposing process, the amplitude spectrum of the voice signal is synthesized according to the decomposing result, and then an enhanced voice signal is restored by combining a phase spectrum of a primary mixed signal. The test can be carried out under the different voice sparsity limitations and temporary continuity limitation, therefore the voice enhancement effect in the music background of the music can be effectively improved by improving the temporary continuity limitation of the background music.

Description

technical field [0001] The invention relates to a speech enhancement method, in particular to a speech enhancement method under music background considering the sparsity of speech signal and the limitation of temporal continuity of background music. In the field of speech analysis or synthesis, audio analysis or processing. Background technique [0002] Due to the interference of background noise, the performance of automatic speech recognition system will drop sharply with the decrease of signal-to-noise ratio. At present, there are good speech enhancement algorithms that can well enhance the speech signal under stationary noise, such as spectral subtraction, adaptive filtering method, minimum mean square error estimation method, etc., so as to improve the recognition rate of automatic speech recognition system under stationary noise . But for the most common non-stationary noise in daily life, there is no good solution to enhance the speech signal in such noise environme...

AI Technical Summary

Problems solved by technology

But for the most common non-stationary noise in daily life, there is no good solution to enhance the speech signal in such noise environment

This also makes the speech recognition effect in the actual environment unsatisfactory.

[0003] As a specific background noise, the music signal often appears together with the target voice signal, such as answering the phone while playing music while driving, or in some entertainment venues that play background music, etc., thereby polluting the pure voice signal and reducing the automatic voice signal. The recognition rate of the recognition system in the background of music

However, due to the non-stationary and spectral features of the music signal and the similar characteristics of the speech signal, it becomes very difficult to extract the speech signal from the music background

[0004] The existing algorithms for extracting speech signals from background music have achieved good results, but most of the algorithms do not fully consider the prior information of speech and background music signals, or only consider the sparsity characteristics of speech signals

On the other hand, most systems are speaker-dependent systems, which require the system to know the identity of the speaker in advance to achieve the best results, which also limits the application scenarios of the system to a certain extent

Images