Audio event detection method, device and equipment

Abstract

The invention discloses an audio event detection method, device and equipment. The method comprises the following steps: generating a spectrogram of an audio event, extracting universality characteristics of an audio corresponding to the audio event according to the spectrogram, and detecting the audio event according to the universality characteristics. By means of the mode, due to the fact thatthe spectrograms have the good characteristics of intra-class similarity and inter-class difference, the difference between different types of audio can be well reflected, and the effect of audio event detection results can be improved.

Description

technical field [0001] The present invention relates to the technical field of audio events, in particular to an audio event detection method, device and equipment. Background technique [0002] Audio events are often defined as classes of recognized audio. Audio event detection and audio classification are consistent and interlinked in nature, and both are pattern recognition techniques. The audio classification is to discriminate the category of the audio event included in the audio clip. [0003] In daily life, what is often encountered is continuous audio stream detection. It is necessary to determine the start and end points of abnormal sounds, accurately determine the category, and give a corresponding response. For the detection of the start and end points of abnormal sounds, the commonly used methods include zero-crossing rate, short-term energy, and double-threshold threshold method. [0004] However, the existing audio event detection schemes generally first fra...

AI Technical Summary

Problems solved by technology

[0004]However, the existing audio event detection schemes generally first divide the training audio into frames and add windows, then extract the feature set for each frame signal, and use the obtained The feature vector is used to train the classifier mode, and then the test audio undergoes the same processing, including frame division and windowing, feature extraction, etc., to obtain a feature set, use the trained model to detect audio events, and output the detection results, but due to the existing The audio event detection scheme does not differentiate audio events between classes, and cannot well reflect the differences between different classes, resulting in a mediocre effect on audio event detection results

Images