Audio processing model training method and device, audio denoising method and device, and electronic equipment

Abstract

The invention relates to an audio processing model training method. An audio processing model comprises a to-be-trained feature dimension reduction model, a to-be-trained audio denoising model and a to-be-trained feature dimension raising model. The method comprises the following steps: acquiring a noise-carrying audio signal and a corresponding pure audio signal; inputting the audio signal with noise into the to-be-trained feature dimension reduction model to obtain an audio feature after dimension reduction; inputting the audio features after dimension reduction into the to-be-trained audiodenoising model to obtain denoised audio features; and inputting the denoised audio features into the to-be-trained feature dimension raising model to obtain denoised audio signals: based on the difference between the pure audio signals and the denoised audio signals, adjusting model parameters of the to-be-trained audio processing model until a model loss value is lower than a preset threshold. The audio denoising accuracy can be improved.

Description

technical field [0001] The present disclosure relates to the field of audio processing, and in particular, to training of an audio processing model, an audio denoising method, device, electronic equipment, and storage medium. Background technique [0002] During the playback of an audio file, the noise in the audio often makes the audio listener feel harsh. At the same time, the noise in audio often affects the expression of information in the entire audio. This makes audio denoising processing play an important role in the field of audio signal processing. [0003] Related technologies often use neural networks to solve the problem of audio signal denoising processing. When training a neural network, the low-dimensional features output by the neural network are often used to directly optimize the loss of the neural network; however, it is often impossible to make a fair comparison between the final output of the audio denoising process and The resulting neural network is...

AI Technical Summary

Problems solved by technology

[0005] The present disclosure provides an audio processing model training, audio denoising method, device, electronic equipment and storage medium, so as to at least solve the problem of low audio denoising accuracy in related technologies

Images