Segmentation clustering method and system for multi-person voice in complex environment

Abstract

The invention discloses a segmentation clustering method and system for multi-person voice in a complex environment. The method comprises the following steps of: acquiring multiple continuous multi-person speaking voice segment audios according to multi-person speaking audios; and normalizing the multi-person speaking voice segment audios according to acoustic features to obtain normalized audios;acquiring multiple sections of to-be-processed audios; extracting voiceprint information characteristics of the multiple sections of to-be-processed audios; acquiring scores among all the to-be-processed audio segments by setting scoring criteria; according to the similarity scores among all the to-be-processed audio segments, acquiring category labels of a plurality of persons through a multi-stage redundant clustering algorithm; and segmenting and clustering the multi-person speaking audios according to the category labels of the plurality of persons. By using the redundant clustering method, the clustering center of a target speaker can be improved to be more dispersed, and the distinction degree is higher. And for an unclear voice segment of the target speaker in a complex environment, a better discrimination capability is realized, so that the classification error of speaker classification in a segmentation clustering task in the complex environment is reduced.

Description

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine

Login to View More

AI Technical Summary

Benefits of technology

The use of this redundancy algorithm improves the accuracy of identifying specific people or things by dividing them into smaller groups based on their similarity between similar sounds called clusters. This helps identify even if there are many different voices speaking at once without being mistakenly identified due to background noise interference from other sources like loudspeakers.

Problems solved by technology

This patented describes an algorithm called Vadelta which helps detect or cluster sounds without being detected incorrectly due to noise interference between them. Additionally, current algorithms either require advanced signal analysis techniques like wavelets or convolutional networks to handle complicated scenarios involving many talker voices simultaneously have limitations because they often result in incorrect results if certain parts were mistakenly identified during the final step. Therefore, there remains a challenge in optimizing speaker separation/clustering in these challenges.

Images