Speaker segmentation method based on long short-term memory deep neural network

Abstract

The present invention provides a speaker segmentation labeling method and device based on long-short-term memory neural network, which is characterized in that the speaker recognition sample labeling model based on long-short-term memory deep neural network is used to detect each speech from the audio to be tested. The appearance and duration of human speech, including: step S1, preprocessing the audio to be tested to obtain audio frame-level features f1 and audio frame-level features f2; step S2, building a speaker recognition sample labeling model based on long short-term memory deep neural network , the speaker sample labeling model includes a speaker switching detection sub-model and a speaker feature modeling sub-model; step S3, respectively training the speaker switching detection sub-model and a speaker feature modeling sub-model; step S4, audio frame-level The feature f1 and the audio frame-level feature f2 are input into the speaker recognition sample labeling model based on the long short-term memory deep neural network to complete the classification record of the speaking time period of each speaker in the audio to be tested.

Description

technical field [0001] The invention belongs to the technical field of computer hearing and artificial intelligence, and relates to a method for speaker switching detection, speech feature space modeling and segmentation labeling in an auditory scene, in particular to a speaker switching detection based on a bidirectional long-short-term memory network model And voice segment labeling method. Background technique [0002] With the rapid improvement of machine learning technology and computer hardware performance, breakthroughs have been made in recent years in computer vision, natural language processing and speech detection and other application fields. Speaker segmentation is a basic task in computer speech processing, and its recording accuracy has been greatly improved. [0003] The task of speaker segmentation annotation can be divided into two key subtasks: speaker switch detection and speech feature space modeling. [0004] Among them, the speaker switching detectio...

AI Technical Summary

Problems solved by technology

However, on the one hand, collecting a large amount of training data is an extremely difficult task, and the increase in the amount of data also leads to a longer model training time; on the other hand, the clustering method used by the existing methods cannot adapt to the generated speaker Issues such as feature space, segmentation and recording accuracy will be affected

Images