Audio segmentation method based on signal energy spike identification

Abstract

The invention relates to an audio segmentation method based on signal energy spike identification. The audio segmentation method comprises the steps of carrying out short-time Fourier transform on aninput audio signal, and converting the input audio signal into a power spectrum matrix; extracting intermediate frequency energy characteristics based on a power spectrum; carrying out peak identification on the extracted intermediate frequency energy characteristics; carrying out error division correction on the signal after peak identification; and outputting a time coordinate of the division point of the audio signal. The audio segmentation method does not need to set a threshold value, does not need to be trained in advance, analysis can be realized on the basis of audio signals in real time quickly and accurately, the method can be deployed at the edge end, other operating parameters do not need to be accessed, and parameter-free dynamic segmentation is basically realized.

Description

technical field [0001] The present application relates to an audio segmentation method based on signal energy peak recognition, which is applicable to the technical field of audio signal processing. Background technique [0002] The main implementation schemes for the pure audio segmentation algorithm are: [0003] 1. A segmentation method based on endpoint detection, such as the Chinese patent application number CN200510061358.6. Utilizing the characteristic that the speaker pauses between speeches, all silent points are detected as potential points where the speaker may change. Such methods are inaccurate due to the fact that silent points are difficult to detect under different signal-to-noise ratio environments. [0004] 2. Model-based segmentation methods, such as Chinese patents with application numbers CN201710512310.5 and CN201811581291.2. Firstly, corresponding models are established for different types of audio segments, and then the model maximum likelihood sel...

AI Technical Summary

Problems solved by technology

[0007] 1. The positioning of the split point is not accurate. The actual rotation process is continuously changing. If the rotation time of each blade is calculated and divided according to the average rotation speed within a certain resolution time range, it can only be roughly uniform. The length of the blade, but the actual rotation process does not necessarily take the same length of time for each blade

Therefore, this method is only suitable for reference and is not suitable as an accurate input for other analysis algorithms;

[0008] 2. Connecting to the real-time speed of fan blades requires high sensor installation. Acquisition of high-precision speed requires additional sensor hardware in the acquisition equipment. The project implementation is difficult and costly, which is not conducive to maintenance, and because the main shaft speed acquisition is in the engine room of the fan position, and the collector is arranged at the base of the tower, too long signal transmission line will cause interference in the collected signal, poor data quality, and seriously affect the segmentation and interpretation

Images