Method for detecting and positioning smoothing processing in voice segment based on autoregression model coefficient

Abstract

The invention provides a method for detecting and positioning smoothing processing in a voice segment based on an autoregression model coefficient. The method comprises the following steps: S1, constructing an original voice set and a smooth voice set; S2, extracting an AR coefficient of the original voice set as an original voice feature set; extracting an AR coefficient of the smooth voice set as a smooth voice feature set; S3, respectively and randomly screening out an original voice feature set sample and a smooth voice feature set sample, and training an SVM support vector machine classifier; S4, selecting to-be-tested voice, framing the to-be-tested voice, and extracting an AR coefficient from each frame of to-be-tested voice signal to serve as a to-be-tested voice feature set; and S5, classifying the to-be-tested voice feature set by using a trained SVM classifier, judging whether the signal is subjected to smoothing processing or not, and if the signal is subjected to smoothingprocessing, positioning the position of smoothing processing. According to the method provided by the invention, frequency information is not needed, the calculated amount in the detection process isreduced, and the accuracy of detection and positioning is improved.

Description

technical field [0001] The present invention relates to the technical field of smoothing processing detection of speech segments, and more specifically, to a method for detecting and locating smoothing processing in speech segments based on autoregressive model coefficients. Background technique [0002] With the continuous development of multimedia technology, people can easily obtain a variety of digital audio, and with the popularization and application of some professional audio editing software such as Audition, people can easily use audio editing software to edit digital audio. Editorial modification. As evidence, digital audio plays an extremely important role in the judicial field. Therefore, it is necessary to detect the authenticity of digital audio. [0003] Smoothing is a common audio post-processing method, which is often used to falsify the edge of digital audio. Therefore, detecting the authenticity of a digital audio speech segment can be assisted by detect...

AI Technical Summary

Problems solved by technology

[0004] At present, there are existing methods for smoothing speech segments using frequency information such as MFCC and other feature detection. These methods are only suitable for detecting longer speech segments. When the speech segment is very short, the frequency information contained in the speech segment is very limited. The above methods Can not achieve a good detection effect; Sun Yat-sen University applied for a patent in 2018 with the publication number CN110060703A "A method for detecting and locating smoothing in speech segments" and Q.Yan, R.Yang and J.Huang in 2019 In the journal "IEEE Transactions on Information Forensics and Security", the research on the topic "Detection of Speech Smoothing on Very Short Clips" was published, all of which analyzed the differential signals of speech segments, using the standard deviation, differential The standard deviation of the high-frequency component of the signal and the standard deviation of the differential signal of the median filter residual are used as features to identify whether the speech segment has been smoothed. It has a high detection accuracy and can effectively detect and locate the linear There are 6 common smoothing operations including filters and nonlinear filters. However, multiple differential calculations and frequency transformations are required, and when the smoothing window is short, good detection results cannot be achieved, and stability not tall

Images