Voice activity detection method in communication-terminal double-microphone denoising system and apparatus thereof

Abstract

The invention discloses a voice activity detection method in a communication-terminal double-microphone denoising system and an apparatus thereof. The method is characterized by firstly, training a neural network: selecting a training sample, extracting a characteristic and acquiring a corresponding voice activity detection result; using the characteristic and the corresponding detection result to train the neural network; secondly, based on the trained neural network, carrying out voice activity detection: using primary and secondary microphones of a communication terminal respectively to collect voice signals with noises to be detected; extracting a characteristic from the collected voice signals with noises, sending the characteristic to the trained neural network and using the neural network to output a voice activity detection result. The characteristic comprises that a sub-band mutual channel energy difference is related to a normalized mutual channel. According to different noise environments, a parameter is adaptively adjusted and the voice activity detection is performed. A problem that an existing voice activity detection method can not adapt to changes of a noise environment so that performance is decreased is solved. Voice activity detection accuracy under a complex noise environment is increased.

Description

technical field [0001] The invention relates to noise suppression, elimination, noise reduction and noise elimination communication technology in voice communication, in particular to a method and device for detecting voice activity in noisy voice signals. Background technique [0002] In the prior art, the technology applied to mobile phone noise cancellation mainly adopts setting a fixed threshold to judge speech and noise. [0003] However, communication terminals, such as mobile phones, are in a very complex noise environment during a call. Technologies that rely on fixed thresholds cannot accurately distinguish voice and noise in complex and variable noise environments. Applying to mobile phone noise cancellation systems will Problems that cause voice distortion to reduce intelligibility. [0004] Voice activity detection is an important part of voice enhancement processing. Voice activity detection can determine the position of the start and end points of the voice fr...

AI Technical Summary

Problems solved by technology

[0003] However, communication terminals, such as mobile phones, are in a very complex noise environment during a call. Technologies that rely on fixed thresholds cannot accurately distinguish voice and noise in complex and variable noise environments. Applying to mobile phone noise cancellation systems will Problems that cause voice distortion to reduce intelligibility

Images