Parallel Feature Extraction System and Method for General Specific Speech in Speech Signal

Abstract

The invention discloses a parallel feature extraction system and method for general specific speech in speech signals. The model mainly includes a speech signal, an emotion recognition model, a voiceprint recognition model and a speech recognition model; the output features obtained by the LLD channel are combined with the The output features obtained by the spectrogram and TEO channel are combined to obtain 1*1024-dimensional emotional features; the spectrogram is fed into the convolutional neural network CNN as input, and the spectrogram and MFCC are fed into Seq2Seq, and the model of two channels Fusion adds attention mechanism to form voiceprint recognition model; 42-dimensional MFCC is used as input, combined with BIMLSTM and Seq2Seq channel for output; spectrogram is combined with Seq2Seq channel for output; model fusion of two channels is added with attention mechanism to form speech recognition Model. The invention has the advantages of high accuracy rate, high integration degree, free choice of scheme and the like.

Description

technical field [0001] The invention relates to the field of signal processing and extraction, in particular to a feature extraction system for speech tasks. Background technique [0002] Speech is the most effective, natural and important form of communication for human beings. Realizing communication between humans and machines through voice requires machines to be intelligent enough to recognize human voices. With the development of machine learning, neural network and deep learning theory, the completion of speech recognition related tasks is gradually improving, which is of great help to the computer to understand the content of speech. Currently, speech recognition tasks mainly involve the following three recognition tasks: [0003] 1. Voiceprint recognition [0004] Voiceprint recognition, also known as speaker recognition, is a form of biometric recognition. It analyzes and processes the continuous speech signal of the speaker to extract discrete speech features, a...

AI Technical Summary

Problems solved by technology

[0009] However, there are many defects or deficiencies in the actual application or design of the above three recognition tasks

For example: Voiceprint recognition, speech recognition, and emotion recognition task models are not universal, input forms are not uniform, there is no universal solution, the integration accuracy is not high, the recognition accuracy rate of a single emotion recognition task is not high, etc.

Images