Song semantic information indexing method and device, equipment, medium and product

Abstract

The invention discloses a song semantic information indexing method and device, equipment, a medium and a product, and the method comprises the steps: encoding audio information in song audio data, and obtaining corresponding encoding information; adopting a plurality of convolution blocks in a shared network of a feature extraction model trained to a convergence state to perform multistage feature extraction on the encoding information in sequence, and obtaining intermediate feature information of deep semantic information of the song audio data; a global branch network of the feature extraction model is adopted to extract global significant features of the intermediate feature information, and a global output feature vector is obtained; equally segmenting the intermediate feature information according to channels by adopting a local branch network of the feature extraction model to respectively extract semantic local features, and obtaining channel output feature vectors; and splicing the global output feature vector and the channel feature vector into a high-dimensional index vector. According to the method, representation learning of the deep semantic information of the song audio data can be realized.

Description

technical field [0001] The present application relates to the technical field of music information retrieval, and in particular, to a method for indexing song semantic information and its corresponding device, computer equipment, computer-readable storage medium, and computer program product. Background technique [0002] With the popularity of short videos, live broadcasts, and radio stations, the amount of cover music is increasing, and the scenes that require music recognition are becoming more and more complex. Compared with the original version, the cover version may have differences or even completely different musical components such as timbre, fundamental frequency, rhythm, speed, harmony, lyrics, singing style, and overall structure. Therefore, cover recognition is a very challenging research work. [0003] There are a variety of technologies related to cover recognition in the prior art, and various prior technologies all have a certain degree of deficiencies, suc...

AI Technical Summary

Problems solved by technology

[0003] There are many technologies related to cover recognition in the prior art, and all existing technologies have certain deficiencies, for example: (1) the traditional technology of listening to songs and recognizing songs based on Landmark can only recognize songs of the same source, Unable to recognize the above-mentioned cover version with certain differentiated information; (2) The traditional humming recognition technology based on melody matching can only recognize clean a cappella / humming, but cannot recognize the above-mentioned cover version with background accompaniment; (3) The traditional cover recognition technology scheme mainly extracts audio features such as Pitch Class Profile (PCP), and then uses algorithms such as dynamic programming to calculate the similarity distance between songs

Due to the diversity of cover versions, the above scheme is only suitable for adapting smaller cover solutions, the accurate recognition rate is low, and the recognition speed is slow, so it cannot be applied to the search of massive music

[0004] Therefore, the technical solutions related to song recognition in the prior art lack universal adaptability, the recognition accuracy is not high, and the recognition efficiency is low. It is necessary to explore more effective technical solutions

Images