Audio noise reduction system and method based on structure of GRU network

Abstract

The invention belongs to the technical field of Internet audio processing, and discloses an audio noise reduction system and method based on a structure of a GRU network. The noise reduction method based on the structure of the GRU network comprises the following steps: enabling input signals to pass through an overlapping window at first, wherein the overlapping window is a Vorbis window, and meets the standard of Princen-Bradley; fundamental tone analysis: removing harmonic noises by using a fundamental tone filter; carrying out feature extraction on audio after fundamental tone analysis; after the feature extraction step is carried out, converting the input signals into data containing N* 42 feature values after feature extraction, wherein N is a sequence length; when RNN noise reduction is carried out, after three layers of GRU processing are carried out on the data, estimating a noise frequency spectrum, and calculating to obtain 22 frequency band gains through spectral subtraction; carrying out gain processing on the original audio signals divided into 22 frequency bands so as to fulfill a noise reduction function. Compared with a traditional audio noise reduction scheme suchas Speex, the audio noise reduction system and method realize better properties to a certain extent.

Description

technical field [0001] The invention belongs to the technical field of Internet audio processing, and in particular relates to an audio noise reduction system and method based on a GRU network structure. Background technique [0002] At present, the closest existing technology: At present, mobile Internet technology has begun to be applied to all aspects of life, and mobile Internet communication technology has replaced mobile phone functions to a certain extent. Using VoIP in the mobile Internet environment can achieve the same functions as traditional telephony, but the charges are cheaper. However, due to the difficulty of guaranteeing network quality in the mobile Internet environment, the communication quality of VoIP is often low, and there are often various interference factors in the communication process. Traditional audio noise reduction solutions often use spectral subtraction to denoise network audio signals. But its effect is not ideal to a certain extent. [...

AI Technical Summary

Problems solved by technology

[0003] To sum up, the problems existing in the existing technology are: the existing audio noise reduction technology has a lot of noise in the communication quality of Internet telephony, which affects the communication experience and limits the development of Internet telephony to a certain extent.

Images