A method and device for extracting a target sound source

Abstract

The invention discloses a method for extracting a target sound source, comprising: acquiring audio data collected by a first number of sound collection devices; determining a second number of sound sources in the audio data; combining the first number with the The second number is compared, and according to the comparison result, the corresponding target blind source separation method is selected to carry out blind source separation on the audio data, and each sound source is obtained; the sound source that has a preset activation word in the various sound sources is selected as the target sound source. In the above extraction method, when the audio data is received, the second quantity of the sound source of the audio data is first determined, and the accurate positioning of the sound source is realized, and then the target sound source is extracted, which avoids the problem in the prior art. In noise and reverberation scenes or different sound sources with short intervals, the positioning results have large errors, and accurate positioning cannot be achieved.

Description

technical field [0001] The invention relates to the technical field of speech extraction, in particular to a method and device for extracting a target sound source. Background technique [0002] Extracting the target sound source of interest in a noisy environment is called the "cocktail party problem". In the actual environment, the propagation of sound is reflected, multipathed, and superimposed, and when it reaches the microphone, it is very different from the source sound source. It becomes extremely difficult to extract the sound of interest from the microphone. [0003] In the prior art, sound source data is collected using sound acquisition equipment, and then SRP and particle filters are used to locate and track the sound source data. Blind speech separation separates the microphone signal according to the sound source orientation information and performs post-filtering to obtain an independent sound source. source. In the above method, in the noise and reverberati...

AI Technical Summary

Problems solved by technology

[0004] In view of this, the present invention provides a method and device for extracting target sound sources, which are used to solve the problem that in noise and reverberation scenes or different sound sources with short intervals, the positioning results have large errors and accurate positioning cannot be achieved. question

Images