Voice sensitive information detecting and filtering method based on unspecified people

Abstract

The invention discloses a voice sensitive information detecting and filtering method based on unspecified people. The method is capable of detecting and filtering real-time voices and voice files. The method includes the steps: carrying out end point detection for an original voice through an improved double threshold end point detection algorithm, extracting the Mel frequency inverted spectral coefficient characteristic of the voice, training an appropriate voice characteristic template through a self-learning time warping algorithm, and storing the template in a database; and then performing end point detection for the original voice through the improved double threshold end point detection algorithm, extracting the MFCC characteristic, comparing the extracted voice characteristic with the template in the sensitive word characteristic template database by combining rough matching with fine matching, detecting sensitive words input to the voice, and filtering the detected sensitive words.

Description

technical field [0001] The invention relates to multimedia content security technology, in particular to a non-specific person-oriented voice sensitive information detection and filtering method. Background technique [0002] With the development of telecommunications network technology, voice applications such as telephone voice and network audio are becoming more and more mature. However, voice communication containing sensitive information and illegal information is not conducive to social harmony and stability and the long-term stability of the country. How to detect sensitive information from massive speech data has become an urgent problem to be solved. [0003] The traditional method is to use artificial listening, which is only suitable for processing a small amount of speech, but it is inefficient and often consumes huge manpower and material resources when performing manual detection of massive speech information, but it is difficult to achieve satisfactory detect...

AI Technical Summary

Problems solved by technology

Although this method uses automatic computer processing to avoid waste of human resources, the speech recognition process is slow, and it is difficult to apply to real-time interactive voice programs such as TV and radio, and voice chat rooms that require high real-time performance.

[0004] To sum up, in the existing technology, non-specific person-oriented speech sensitive information detection and filtering methods have great limitations, high missed detection rate and false detection rate, and it is difficult to meet real-time requirements

Images