Sound event detection method based on full convolutional network

Abstract

The invention discloses a sound event detection method based on a full convolutional neural network, and mainly solves the problems of low multi-audio event detection precision and high time complexity of an existing network. According to the implementation scheme, the method comprises the following steps of 1) performing Mel cepstrum feature extraction on an audio stream to obtain time-frequencyfeature graphs of the audio stream, and forming a training data set by using the time-frequency feature graphs; 2) establishing a full convolution multi-audio event detection network composed of a frequency convolution network, a time convolution network and a decoding convolution network from top to bottom; 3) training the full convolution multi-audio event detection network by using the data set; and 4) inputting the audio stream to be detected into the trained full convolution multi-audio event detection network for multi-audio event detection to obtain the category of the audio event and the existing starting and ending time. Simulation results show that compared with an existing network 3D-CRNN with the highest precision, the precision of the method is improved by 2%, the operation speed is improved by about 5 times, and the method can be used for safety monitoring.

Description

technical field [0001] The invention belongs to the technical field of event detection, in particular to a multi-audio event detection method, which can be used for security monitoring. Background technique [0002] Audio event detection refers to locating sound events of interest in an audio stream and classifying them correctly. Audio information is more practical than image information in the fields of disaster relief, gunshot monitoring, etc. In addition, audio information can also assist video information to complete video surveillance, search tasks, etc., so it is of great practical significance to realize audio event detection. [0003] Sound event detection on real audio streams obtained in real life has always been a very challenging task, because first of all, in reality, different events are highly concurrency, and different sound events often occur at the same time, which requires audio events The detection system can identify multiple audio events from the alia...

AI Technical Summary

Problems solved by technology

However, because the RNN network cannot process data in parallel, the time complexity of the algorithm is too high, and the network training time is too long

[0007] Although the above-mentioned existing methods can complete the multi-audio event detection task to a certain extent, there are still the following problems: 1) CRNN network captures time series information through LSTM and other RNN networks, but the maximum dependency length that LSTM can effectively capture is Only between 30 and 80, and it is a single scale information

This makes it difficult for RNN-like networks to complete localization and classification at the same time, and the accuracy is low when performing audio event detection tasks with different durations

2) The RNN network is difficult to process in parallel, and the algorithm time complexity is high, resulting in too long network training time

Images