Kernel-function-based dimension reduction method of audio feature signal

Abstract

The invention, which belongs to the technical field of audio signal processing, relates to a kernel-function-based dimension reduction method of an audio feature signal. Dimension reduction processingis carried out on a feature parameter of an audio signal; the required dimension reduction effect is realized while the audio feature information amount is not abandoned; and then last dimension reduction data are displayed visually and comparative analysis is carried out on the last dimension reduction data and results obtained by other audio feature parameter dimension reduction methods. According to the invention, for dimension reduction of the audio feature parameter, dimension reduction processing is mainly carried out on a linear prediction coefficient, a linear prediction cepstral coefficient and a Mel frequency cepstral coefficient in an audio coefficient domain and the data results after dimension reduction are displayed visually. The audio feature dimension reduction processingcan be applied to broadcast signal monitoring and quick identification processing of audio signals. The algorithm is simple and the nonlinear kernel function is used for expressing a mapping relationship between Gaussian observation space and hidden space, thereby overcoming defects of usage range limitation and poor dimension reduction effect of the linear mapping method.

Description

technical field [0001] The invention relates to a dimensionality reduction method of an audio feature signal based on a kernel function, and belongs to the technical field of audio feature signal processing. Background technique [0002] In order to realize the management and control of wireless audio broadcasting, and carry out safe and efficient real-time monitoring and screening of audio broadcasting, the rapid processing of audio information is related to the process speed of the entire process, and the dimensionality reduction processing of audio characteristic signals is the core of audio information processing. Efficiency and credibility must also become problems to be solved urgently. At present, most of the dimension reduction methods for audio feature signals mainly include local preservation projection method, multi-dimensional scaling method, local linear embedding method, principal component analysis method, etc. Most of these dimensionality reduction algorithm...

AI Technical Summary

Problems solved by technology

Most of these dimensionality reduction algorithms are highly complex, to discard part of the characteristic signals to achieve the purpose of dimensionality reduction, which will cause unpredictable errors in actual engineering applications. The present invention is proposed for the above drawbacks

Images