Environment Adaptive Speech Enhancement Algorithm Based on Attention-Driven Recurrent Convolutional Network

Abstract

The invention discloses an environment-adaptive speech enhancement algorithm based on attention-driven circular convolution network, comprising the following steps: step 1, selecting a speech enhancement task database, and preparing input data; step 2, extracting the amplitude information of the speech and the environment Information, wherein the environmental information of the speech is extracted by using the weighted prediction error method (WPE), and the amplitude information of the speech is mainly the spectrogram information extracted by Fourier transform; Step 3, the construction and training of the depth model; Step 4, the speech Reconstruction, converting the speech amplitude predicted in step 3 into a speech waveform. The present invention considers the environmental information of the voice, and improves the environmental adaptability and algorithm robustness of the algorithm in different environments; in terms of preserving the real voice signal, the present invention incorporates the attention mechanism to build an attention-driven circular convolution network, which is more Accurately characterizes the temporal context information of speech, effectively improving the performance of speech enhancement.

Description

technical field [0001] The invention belongs to the technical field of speech enhancement, in particular to an environment-adaptive speech enhancement algorithm based on an attention-driven circular convolution network. Background technique [0002] With the popularization of smart devices and the rapid development of speech recognition technology, speech processing technology has attracted more and more public attention. In a common near-field environment (the speaker is relatively close to the microphone), the performance of speech recognition has reached over 95%, and many speech recognition and speech synthesis technologies have been commercialized. However, in the far-field environment (the speaker is far away from the microphone), there is often the influence of reverberation and various background noises, and the performance of speech recognition drops sharply. In the far-field environment, since the speaker does not need to hold a microphone or wear a microphone dev...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem that the existing speech enhancement model is difficult to adapt to different noise environments, the present invention proposes an environment-adaptive speech enhancement algorithm based on attention-driven circular convolution network, thereby improving the environmental adaptability of the algorithm in different environments Algorithmic Robustness

Images