Mixed sound signal separation method, device, electronic device and readable medium

Abstract

The present application relates to a separation method, device, electronic equipment and readable medium of a mixed sound signal. The method includes: extracting mixed sound feature data from the mixed sound signal; inputting the mixed sound feature data into a mixed sound coding model to obtain a first latent variable and a second latent variable, the first latent variable characterizes vocal features, and the second The two hidden variables represent the sound characteristics of the accompaniment; the first hidden variable and the second hidden variable are respectively input into the human voice decoding model and the accompaniment decoding model to obtain the human voice characteristic data and the accompaniment sound characteristic data; and based on the human voice characteristic data and the accompaniment sound Characteristic data get vocals and accompaniment. The signal-to-noise ratio of human voice and accompaniment sound obtained by this method is low.

Description

technical field [0001] The present application belongs to the application field of computer software, in particular to a separation method, device, electronic equipment and readable medium of a mixed sound signal. Background technique [0002] General pop music is composed of vocals and accompaniment superimposed and mixed. It is a challenging task to separate the mixed music into vocals and accompaniment (accompaniment separation). Acoustic partner separation plays an important role in music editing and music retrieval. Improvements in the performance of the Acoustic Separation model can greatly improve the performance of subsequent processing pipelines. [0003] The current mainstream sound companion separation model is an end-to-end deterministic model, which calculates the mask (mask) of each sound source in the time-frequency diagram, and then multiplies the mask by the time-frequency diagram of the mixed sound to obtain the time-frequency of the separated sound sourc...

AI Technical Summary

Problems solved by technology

Although the sound source signal obtained by this end-to-end model separation has a high signal-to-noise ratio, it is almost impossible for the separated sound source signal to be clean, and it will be more or less doped with other residual sound source signals

Although these residual interferences are weak, they have a very serious impact on subsequent steps such as lyrics segmentation and song evaluation.

Images