Speech enhancement method and device based on local attention mechanism and medium

Abstract

The invention discloses a speech enhancement method based on a local attention mechanism, and the method comprises the steps: collecting a to-be-processed speech, carrying out the speech synthesis ofthe to-be-processed speech, and obtaining a to-be-enhanced speech; performing feature extraction on the to-be-enhanced speech to obtain multi-dimensional features of the to-be-enhanced speech; establishing an RNN model based on a local attention mechanism, and inputting the multi-dimensional features into the RNN model; training the RNN model to obtain a speech enhancement model; calculating the multi-dimensional features according to a speech enhancement model to obtain a time-frequency masking factor, and obtaining the enhanced speech spectrum of the to-be-enhanced speech according to the time-frequency masking factor and the frequency spectrum of the to-be-enhanced speech; and obtaining a speech enhancement waveform according to the enhanced speech spectrum. According to the speech enhancement method and system based on the local attention mechanism, the speech enhancement effect can be effectively improved, and the speech enhancement stability and reliability can be effectively improved.

Description

technical field [0001] The present invention relates to the technical field of speech processing, in particular to a speech enhancement method, device and medium based on a local attention mechanism. Background technique [0002] Speech enhancement is an important topic in the field of speech processing. It refers to the technology of extracting useful speech signals from the noise background to suppress and reduce noise interference when the speech signal is interfered or even covered by various noises. With the development of related applications such as smart speakers and human-computer interaction, in an ideal environment, algorithms such as speech recognition and speaker recognition have a high accuracy rate, but in actual scenarios, the existence of noise makes these speech applications accurate. The intensity is greatly reduced, so reducing the interference of noise on the speech signal and enhancing the speech signal are urgent problems to be solved. [0003] At pre...

AI Technical Summary

Problems solved by technology

[0005] To a large extent, it depends on the estimation of noise, and these methods are applicable to few scenarios, and fail to fully consider the speech characteristics in each scenario, which inevitably introduces distortion, resulting in poor speech enhancement effect

Images