Single channel-based non-supervision target speaker speech extraction method

Abstract

The embodiment of the invention discloses a single channel-based non-supervision target speaker speech extraction method comprising a teacher language detection step and a teacher language model training step; the teacher language detection step comprises the following parts: obtaining speech data from a classroom recording; processing speech signals; speech segmentation and modeling, the speech segmentation comprises steps of segmenting the classroom speech at equal length, aiming at each segment of speech and extracting corresponding MFCC features, and building each segment speech GMM modelaccording to the MFCC features; teacher speed detection, calculating the similarity between the GMM model of each segment speech except for teacher speech types and a GGMM, tagging the GMM models smaller than a set threshold as teacher speech types, thus obtaining the final teacher speech types; the teacher language GGMM model training step comprises the following parts: clustering the speech dataobtained in S3; obtaining an initial teacher speech type, and extracting the GGMM model according to the initial teacher speech type. The method can effectively improve the system adaptability and intelligence in real applications, thus laying foundation for following applications and researches.

Description

technical field [0001] The invention relates to a voice extraction method, in particular to a single-channel, unsupervised target speaker voice extraction method in a complex multi-speaker situation. Background technique [0002] The assurance of the quality of education is the key to our education at all levels. In improving the quality of education, improving the quality of teaching, especially the quality of classroom teaching should be the top priority. However, the current traditional method is based on manual (peer) on-site observation and evaluation. Although this type of method can play a certain role, it does not have universal operability and universal objectivity. The reasons are: It is difficult for a teaching authority to inspect the classroom, make evaluations and give suggestions all the time, which will inevitably bring a heavy burden to teaching management and is unnecessary. Furthermore, traditional on-site observation and evaluation cannot objectively ev...

AI Technical Summary

Problems solved by technology

[0015] MFCC coefficients are clustered, and MFCC extracts corresponding features based on framing speech. For longer speech segments, such as 40-minute classroom recordings, the amount of calculation will be large, and the clustering accuracy cannot be guaranteed.

[0016] 4. The article > is based on CASA for speech separation, simulating the human ear for speech separation, but the characteristics of the model human ear are difficult to select

[0018] 6. The influence of noise still exists in the results of single-channel speech separation. The above speech separation methods seldom further denoise the speech separation results and purify and separate the speech signals.

Images