Sound event detection and positioning method based on deep learning

Abstract

In an overlapped sound event detection task, sometimes, extracted global features cannot accurately detect and locate sound events of overlapped parts. In view of the fact that short-term and long-term sequence features of a sound event related to a context are obtained by utilizing a Gated Recurrent Unit (GRU) based on a multi-scale spatial channel squeeze excitation convolutional network and the GRU, the invention provides a sound event detection and positioning model based on multi-scale spatial channel squeeze excitation (MscSE). The model, a baseline model and a residual network model are subjected to a contrast experiment in a public data set DCASE2020Task3. According to the optimal results, the detection ER is 0.59, the F1 score is 50.7%, the positioning error DE score and the DE_F1 score are 15.8% and 70.3% respectively, the F1 score is 2%-5% higher than that of other models, and the ER is also lower than that of other models. Therefore, compared with a single-scale model, the squeeze excitation model based on multiple scales is improved in sound event detection and positioning performance.

Description

technical field [0001] The invention relates to a sound event detection and positioning method based on a multi-scale spatial channel squeeze excitation model, belonging to the field of audio detection. Background technique [0002] The goal of Sound Event Location and Detection (SELD) is to detect the acoustic content contained in the audio information, identify the sound event tags and the corresponding time and space locations. SELD is usually used to detect and locate abnormal sounds such as "gunshots" and "screams". It is not only a supplement to video or image detection, but sound usually consumes less computer resources than video storage and processing. SELD has always been a research hotspot, and has been applied in many fields, such as: multimedia, transportation, medical care, safety detection system, etc. [0003] The Hidden Markov model (HMM) was used to detect relatively static events in the early stage. Viterbi decoding produces a sequence of sound events an...

AI Technical Summary

Problems solved by technology

[0009] Aiming at the problem that single-scale convolution cannot fully extract features, resulting in the inability of global features to accurately detect and locate sound events in overl

Images