Multi-speaker Speech Separation Method Based on Voiceprint Features and Generative Adversarial Learning

Abstract

The present invention provides a multi-speaker speech separation method based on voiceprint features and generation confrontation learning to solve the problem of inaccurate and pure speech separation in the prior art. The multi-speaker speech separation method is to mix the audio data of the target speaker, other irrelevant speakers, and noise to obtain an initial mixed training corpus, and to extract the voiceprint features from the pure training corpus of the target speaker and the separation result of the initialization generator, Complete the training of the discriminator; after the parameters of the discriminator are solidified, the training of the generator is completed; the generator with parameter solidification separates the target speaker's speech from the speech to be separated through generative confrontation learning. The invention utilizes generative adversarial learning to generate samples similar to the target, and continuously approaches the output distribution through the generative adversarial network, which reduces the distribution difference between speech data and real target speaker training data in a multi-speaker interference environment, and realizes target speaker Tracking and identification of audio.

Description

technical field [0001] The invention belongs to the field of speech recognition, in particular to a multi-speaker speech separation method based on voiceprint features and generation confrontation learning. Background technique [0002] Automatic Speech Recognition (ASR) is to convert the vocabulary content in human speech into computer-readable input, and use computers to recognize human language. As a way of communication between humans and computers, it is regarded as The basic means of future technology interaction. When people speak in different environments, there will be different interferences. To accurately identify the language of the target speaker, it is necessary to separate the collected audio information. Speech separation includes speech enhancement, multi-speaker separation, and reverberation, among which multi-speaker separation is the most common. For example, in an instant inter-interpretation system for intelligent conferences, on the one hand, when th...

AI Technical Summary

Problems solved by technology

However, the deep model based on spectral mapping has high complexity and strong modeling ability, but its generalization is heavily dependent on the data set. If the amount of data is insufficient, the learned spectral mapping relationship is not robust enough; in addition, feature selection is usually general feature, the speech separation method based on spectral mapping fails to effectively combine the auditory selection characteristics of the human ear and the voice characteristics of different speakers, and the effect of speech separation is not good

Images