Noise detection method and device

Abstract

An embodiment of the invention discloses a noise detection method and device. The method comprises the steps as follows: a to-be-processed audio signal is obtained, and a power spectrum (omega) of an audio frame in the audio signal is computed; omega is the frequency of 2pi*power spectrum; autocorrelation Corr(tau) is computed according to the power spectrum of the audio frame, and tau is a time value; an enhanced correlation spectrum Ecorr (tau) is computed on the basis of the autocorrelation Corr(tau); the maximum value Max (ECorr) in Ecorr (tau) is obtained, and the audio frame is determined as noise if Max (ECorr) of the continuous predetermined number of audio frames is smaller than a first threshold; or, tau corresponding to Max (ECorr) is obtained, and the audio frame is determined as noise if tau corresponding to Max (ECorr) is not in the preset threshold range. Noise is identified on the basis of the enhanced correlation spectrum Ecorr (tau) and can be separated from music and human sound, so that a basis is provided for denoising.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a noise detection method and device. Background technique [0002] Network applications for live broadcasting through mobile phones are gradually popularized, but there are major differences in audio signals during live broadcasting and voice calls. There may be singing or performances during the live broadcast, and there may also be music accompaniment or live accompaniment. [0003] Noise reduction technology will be used in web real-time communication (Web Real-Time Communication, WebRtc) technology, as follows: WebRtc technology uses spectrum flatness parameters and the degree of spectrum change between adjacent frames to calculate the speech / noise probability of each frequency point , then update the noise spectrum, and finally remove the noise by Wiener filtering. [0004] However, WebRtc performs noise reduction processing for speech. When there is music in the backgro...

AI Technical Summary

Problems solved by technology

[0004] However, WebRtc performs noise reduction processing for speech. When there is music in the background sound, especially music fragments whose spectrum does not change (such as long notes of bowed instruments), the noise spectrum will be updated by mistake to suppress this piece of music. , causing damage to the music

Although ordinary autocorrelation detection can detect the correlation peak of music, because most of the environmental noise is pink noise (Pink noise), the correlation peak of music is not obvious in the autocorrelation spectrum of pink noise, so it is rarely used Autocorrelation Spectrum Differentiates Music from Noise

Images