Deep-learning-technology-based automatic accent classification method and apparatus

Abstract

The invention discloses a deep-learning-technology-based automatic accent classification method and apparatus. The method comprises: mute voice elimination is carried out on all accent voices in a training set and mel-frequency cepstrum coefficient (MFCC) feature extraction is carried out; according to the extracted MFCC feature, deep neural networks of various accent voices are trained to describe acoustic characteristics of various accent voices, wherein the deep neural networks are forward artificial neural networks at least including two hidden layers; probability scores of all voice frames of a to-be-identified voice at all accent classifications in the deep neural networks are calculated and an accent classification tag with the largest probability score is set as a voice identification tag of the voice frame; and the voice classification of each voice frame in the to-be-identified voice is used for carrying out majority voting to obtain a voice classification corresponding to the to-be-identified voice. According to the invention, context information can be utilized effectively and thus a classification effect better than a traditional superficial layer model can be provided.

Description

technical field [0001] The invention relates to a method for de-muting speech with an accent and extracting features, a deep neural network modeling, parameter selection, training and inference method, and accent classification. Background technique [0002] Accents in speech are divided into native accents and foreign accents. Native accent refers to the accent produced by the speaker who is influenced by the dialect in his native language when he pronounces it in his native language. Foreign language accent refers to the accent produced by the speaker who is influenced by the pronunciation of the mother tongue when speaking in a non-native language. In the present invention, we mainly aim at the classification problem of mother tongue accent in Chinese. [0003] Commonly used methods for classifying Chinese accented speech mainly fall into two categories: text-dependent methods and text-independent methods. [0004] The text-related accent classification method refers t...

AI Technical Summary

Problems solved by technology

Model Although discriminative training methods (discriminative training methods) of the GMM model such as the minimum classification error criterion (MCE) have been used to suppress the confusion area and also increase the discriminative ability of the accent model, the performance of the text-independent accent classification method is still low. needs to be further improved

Images