Method for separating single-channel mixed voice based on compressed sensing and K-SVD

Abstract

The invention relates to a method for separating single-channel mixed voice based on compressed sensing and kernel singular value decomposition (K-SVD), which comprises the following steps of: constructing a universally applicable overcomplete dictionary, namely a K-SVD dictionary for each of three types of man-man, man-woman and woman-woman mixed training voice by using a K-SVD algorithm through mixed training voice frames; making a signal sparse under the dictionary while a reconfiguration error is in a certain range; on the basis of the constructed K-SVD dictionary, starting from the similarity of compressed sensing observation and single-channel mixed voice expressions, separating the single-channel mixed voice by using a l0-norm optimization-based signal reconfiguration algorithm in a compressed sensing theory; solving the estimation of sparse representation of each source voice frame under the K-SVD dictionary on the basis of the expression of each single-channel mixed voice frame, and reconfiguring each separated voice frame through the estimation of the sparse representation and the K-SVD dictionary; and sequentially connecting the separated voice frames to acquire a separated voice signal.

Description

technical field [0001] The present invention relates to a special category of speech enhancement—speech separation, in particular to a single-channel hybrid speech separation method based on compressed sensing and K-SVD, which belongs to the technical field of speech signal processing. Background technique [0002] Speech is the most convenient, direct and commonly used communication method for human beings. However, in the actual environment, people will inevitably be disturbed by the surrounding noise while acquiring speech signals. These disturbances will affect the performance of speech processing systems (such as speech recognition systems) on the one hand, and affect the performance of human ears on the other hand. Perception and understanding of speech. Therefore, speech enhancement is particularly necessary. Speech separation is a special kind of speech enhancement method. Its noise objects are generally speech-like noises that are difficult to process. It is based...

AI Technical Summary

Problems solved by technology

Although the current research on single-channel mixed speech separation has achieved certain results, the overall complexity of the algorithm is relatively high, and the performance varies greatly despite the difference in the source speech signal. In addition, there are special requirements for the training data in the training phase, so Overall, the practicability is not strong and needs to be improved for specific application

Images