Acoustic Scene Classification Method Based on Network Model Fusion

Abstract

The invention discloses a sound scene classification method based on network model fusion, and constructs a variety of different input features by means of channel separation and audio cutting, and extracts gamma-pass filter cepstral coefficients and plum The spectral features and their first-order and second-order differences are used as input features, and a variety of corresponding convolutional neural network models are trained respectively. Finally, the support vector machine stacking method is used to realize the final fusion model. The present invention uses methods such as channel separation and audio cutting to extract highly recognizable audio input features, constructs a convolutional neural network of single and dual channels, and finally generates a unique model fusion structure, which can obtain more abundant and three-dimensional information , which effectively improves the classification recognition rate and robustness of different acoustic scenes, and has a good application prospect.

Description

technical field [0001] The invention relates to the technical field of acoustic scene classification, in particular to an acoustic scene classification method based on network model fusion. Background technique [0002] The sound scene classification technology is to use computing means to complete the classification of sound scenes according to the information contained in different sound scenes. This technology is of great significance in improving the automation of machines, allowing machines to automatically perceive environmental features, retrieve audio content, and improve the performance of multimedia electronic products. [0003] The features used in traditional acoustic scene classification mainly include: features such as zero-crossing rate and energy in the time domain or features in the frequency domain and cepstrum domain. Commonly used classification methods include: simple threshold judgment method, Gaussian Mixture Model (Gaussian MixtureModel, GMM) method,...

AI Technical Summary

Problems solved by technology

[0007] 2) The recognition ability of the pattern recognition algorithm is greatly affected by the change of the environment, and the robustness is poor;

[0008] 3) Traditional classifiers have weak classification ability and no learning ability

[0009] In addition, a video-based event detection method used in the prior art, due to adverse factors such as insufficient light, dim environment, and too much floating dust in the air, the returned video image is blurred, which is likely to cause misjudgment and missed judgment, as well as acoustic scene classification. , problems with low recognition rate and poor robustness

Images