Teaching checking-in method based on sound recognition

Abstract

The invention discloses a teaching checking-in method based on sound recognition. A processor uses an acoustical model deep neural network in voice recognition to classify sound signals of students; the speaker information is extracted and is used for extracting DNN i-vector for answering student voice voiceprint recognition; the recognized voiceprint of the answering students is subjected to answering matching; five sound pickers are used for space positioning on the sound sources of the answering students; whether the students talks without explicit approval or not is judged according to thesound source position, the sound source direction and the sound source sound volume; the goals of checking-in and class rule monitoring are achieved. In addition, the problems that a base sound period coefficient is difficult to be used for distinguishing unvoiced sound and low-level voiced sound, and the MFCC parameter has poor noise-resistant performance are solved; at the voiceprint recognition stage, the CMM is used for training and recognizing the specific people voiceprint; under the environment noise interference condition, the small sound source position estimation deviation is realized.

Description

technical field [0001] The invention belongs to the field of speech recognition, in particular to a teaching attendance method based on speech recognition. Background technique [0002] Although the current attendance management system has made significant progress compared with the previous sign-in method, because it still relies on wired data transmission, it needs to be completed by more complicated processes such as layout and wiring and line port connection, and this working method is not only time-consuming and labor-intensive. Raw materials are wasted, and line management and data analysis must be done manually. Therefore, it is increasingly urgent for everyone to be able to wirelessly transmit and receive attendance information and realize a low-cost, high-efficiency, and error-free attendance management system. Need, attendance system based on voiceprint information has appeared on the market, but the voiceprint recognition rate is not high in complex classroom voic...

AI Technical Summary

Problems solved by technology

[0002] Although the current attendance management system has made significant progress compared with the previous sign-in method, because it still relies on wired data transmission, it needs to be completed by more complicated processes such as layout and wiring and line port connection, and this working method is not only time-consuming and labor-intensive. Raw materials are wasted, and line management and data analysis must be done manually. Therefore, it is increasingly urgent for everyone to be able to wirelessly transmit and receive attendance information and realize a low-cost, high-efficiency, and error-free attendance management system. Need, attendance system based on voiceprint information has appeared on the market, but the voiceprint recognition rate is not high in the complex classroom voice environment, and the existing voice attendance system does not have the function of monitoring classroom discipline in class

Images