Mobile voice recording device source identification method based on stacked auto-encoding network

Abstract

The invention belongs to the technical field of digital audio data processing, and discloses a mobile voice recording device source identification method based on a stacked auto-encoding network. Themethod comprises first extracting the RASTA-MFCC feature of a pure voice segment and training a GMM-UBM model; then extracting the RASTA-MFCC feature based on the voice segment of a specific device source, adjusting the parameters of the GMM, and extracting the features of the relevant logarithmic spectrum of the voice segment of the specific device source; finally training a deep auto-encoding network by using the extracted features to achieve an automatic identification and classification requirement. The method can determine the source of digital audio in the fields of justice, news, intellectual property, scientific discovery and the like, and verifies the authenticity and the integrity of digital audio data. In the fields of voice recognition and speaker recognition, the method can separately detect the device channel information of training and testing voice, and establishes a channel mapping function between the training and test voice, thereby solving the channel mismatch.

Description

technical field [0001] The invention belongs to the technical field of digital audio data processing, and in particular relates to a source identification method of a mobile recording device based on a stacked self-encoding network. Background technique [0002] At present, the existing technologies commonly used in the industry are as follows: [0003] In recent years, with the development of information technology, editing and tampering of digital audio data has become very easy, and digital audio forensics technology has also received more and more attention, especially in the judicial field, which has important application requirements. Recording device source identification is an audio forensics technology that determines the recording device by analyzing the acquired digital voice signal, and can verify the originality, authenticity, and integrity of the audio. In fact, the source recognition technology of recording equipment has emerged in the famous "** incident". ...

AI Technical Summary

Problems solved by technology

[0007] In the prior art, the ENF signal cannot be extracted from the voice signal based on the mobile terminal device;

[0008] The data based on the speech signal of the silent segment is too small to build a model and the representation is not strong;

[0009] Decision classification accuracy based on shallow network is not high

Images