Audio training or recognizing method for intelligent dialogue voice platform and electronic equipment

Abstract

The invention discloses an audio training or recognizing method and system for an intelligent dialogue voice platform and electronic equipment. The method comprises the steps of receiving audio data;extracting the identity vector characteristics of the audio data, and preprocessing the identity vector characteristics; wherein the preprocessing process comprises extracting Mel Frequency Cepstrum Coefficients (MFCCs) of each frame of audio of the audio data; extracting the identity vector characteristics of the formed 2L+1 frame MFCCs characteristics after the L frames are expanded in the context of the MFCCs characteristic of each frame of audio, so as to generate an on-line identity vector characteristics of each frame of audio; splicing the online identity vector characteristics and theMFCCs characteristics of each frame of audio according to the frame level, so as to generate audio characteristics; and inputting the preprocessed audio characteristics into a gaussian mixture model system based on a background speaker model, so as to carry out audio training or recognition processing. The identity and the speaking content of the speaker can be matched at the same time, and the recognition rate is higher.

Description

technical field [0001] The invention belongs to the technical field of speaker recognition, and in particular relates to an audio training or recognition method, system and electronic equipment for an intelligent dialogue voice platform. Background technique [0002] Speaker verification is the task of identifying whether a target speaker speaks with a test utterance. According to the text content of the test utterance, speaker verification can be divided into two categories: text-dependent and text-independent. For text-dependent tasks, the content of the test and target utterances are constrained to be the same, while text-independent tasks do not have such a constraint. [0003] Over the past decades, various frameworks have been proposed for speaker verification tasks. Currently, the commonly used technical frameworks for speaker verification are: [0004] 1. ivector-PLDA system (Probabilistic Linear Discriminant Analysis, linear discriminant analysis in probabilistic...

AI Technical Summary

Problems solved by technology

However, the text-related speaker verification task requires matching the identity of the speaker and the content of the speech, so the recognition rate of the ivector-PLDA system in the text-related speaker verification task is not ideal; the GMM-UBM system can be used for the speaker to a certain extent identity information modeling, but compared with the ivector-PLDA system, the Gaussian mixture model of the target speaker obtained by matching on the Gaussian mixture model of the background speaker by using a small number of registration sentences cannot describe the speaker’s identity information well, so The accuracy of the system on the speaker verification task is limited; the online ivector combined with the DTW scoring method considers extracting the frame-level ivector as a feature, but its scoring model is too simple, only doing simple pattern matching on the voice content, and the system is reckless Poor stickiness, the system cannot guarantee the ideal recognition effect for tone changes, channel changes, etc.

The above schemes all have the defect of low recognition accuracy for text-related speaker verification tasks

Images