Audio signal conversion method and device

Abstract

The present invention provides a method and device for converting audio signals. After acquiring multiple audio signals, a relationship graph is constructed according to the similarity between every two audio signals calculated in advance, wherein the relationship graph includes a plurality of nodes and connections between nodes. There is a one-to-one correspondence between nodes and audio signals, and each edge of the relationship graph corresponds to a weight, which is determined according to the similarity between the audio signals corresponding to the two nodes connected by the corresponding edge; after obtaining the relationship graph, the relationship graph Execute the graph embedding algorithm to obtain a set of audio vectors corresponding to the relationship graph. The set of audio vectors includes a plurality of audio vectors corresponding to the audio signals one-to-one. The dimension N of the audio vectors is a preset positive integer. In the solution provided by the present invention, the dimension of the audio vector obtained after processing the relationship graph through the graph embedding algorithm can be preset and has nothing to do with the duration of the audio signal, so it can avoid the audio vector with too large dimension, which is convenient for follow-up The audio vectors are used and analyzed.

Description

technical field [0001] The invention relates to the technical field of signal processing, in particular to an audio signal conversion method and device. Background technique [0002] Existing short video playback websites will recommend corresponding short videos according to user preferences. By converting the audio signal of the short video into the corresponding audio vector, the matching relationship between the audio of the short video and the user can be effectively established, thereby improving the accuracy of short video recommendation. [0003] An existing method of converting an audio signal into an audio vector is to split the audio signal into multiple segments, each segment containing only one vocabulary, and then convert each segment based on the existing method of converting vocabulary into vectors Converted into a vector, the vectors corresponding to all the fragments are combined to form the audio vector corresponding to the audio signal. [0004] The pro...

AI Technical Summary

Problems solved by technology

[0004] The problem with the prior art is that since the segment corresponding to each vocabulary needs to be converted into a vector of a specific dimension, if an audio with a long duration and a large number of vocabulary is to be converted, a large dimension vector will be generated. The audio vector not only increases the calculation amount of the conversion process, but also is not convenient for the subsequent analysis and use of the audio vector

Images