Audio data structured conversion method based on working site

Abstract

The invention discloses an audio data structured conversion method based on a working site, relates to the technical field of audio data, and solves the technical problem of poor precision of an existing audio analysis product. The method comprises the following steps: acquiring an original audio in an electric power working site environment; performing feature extraction on the original audio by using a Mel-frequency cepstrum coefficient (MFCC) to obtain MFCC features; processing the original audio by using a Dirichlet process Gaussian mixture model (DPGMM) to obtain a DPGMM posterior graph; connecting the DPGMM posterior graph with the MFCC feature to serve as an audio structuralized enhancement feature of the original audio, and obtaining a voice text; performing multi-label classification on the voice text by using a Catboost algorithm to obtain multi-label classification information; and storing the multi-label classification information into a database so as to facilitate subsequent keyword retrieval and deeper audio analysis. Feature extraction is carried out on the audio data through the MFCC-DPGMM, audio file multi-label processing is carried out through the Catboost algorithm, structured processing of the audio data is achieved, and the analysis precision is good.

Description

Technical field [0001] The present invention relates to the field of audio data, and more particularly, it relates to an audio data structure-based conversion method based on the job site. Background technique [0002] In the power industry, with the energy internet, smart grid, the construction and development of electricity networks, the various network topologies become more complicated. Especially in the intelligent monitoring of power operations, audio data such as field voice is widely collected, so massive audio data is also produced. However, the audio collection and analysis system at the current power operation site only implements data acquisition, massive audio needs to consume a lot of artificial processing and analysis, and there is an unable to intelligently audio data mining, unable to effectively express and manage and efficiently retrieval. How to extract structured information and content semantics in audio is the key to audio information depth processing, cont...

AI Technical Summary

Problems solved by technology

Although some audio-based intelligent analysis products have appeared in recent years as a powerful supplement to manual screening analysis, they lack standards for large-scale application, and the analysis accuracy is poor, so the role is very limited

Images