Audio retrieval method and device and medium

Abstract

The invention discloses an audio retrieval method and device and a medium, and the method comprises the steps: obtaining a to-be-retrieved audio, and extracting the voiceprint features of the to-be-retrieved audio; matching a target voiceprint feature from a preset voiceprint feature library based on the voiceprint feature; determining a target singer corresponding to the target voiceprint feature; screening retrieval audio features of each song of the target singer from a preset retrieval audio feature library to obtain a retrieval audio feature set corresponding to all songs of the target singer; and based on the retrieval audio feature set, retrieving a target song corresponding to the to-be-retrieved audio to obtain a retrieval result. Therefore, the voiceprint features of the to-be-retrieved audio are firstly extracted, then the target singer corresponding to the voiceprint features is matched, and then the retrieval audio feature range is narrowed in the retrieval audio feature set corresponding to all songs of the target singer, so that the retrieval amount is reduced, the features of other singers are excluded, the time consumption and resource overhead of audio retrieval can be reduced, and the retrieval accuracy is improved.

Description

technical field [0001] The present application relates to the technical field of audio retrieval, in particular to an audio retrieval method, device and medium. Background technique [0002] In response to the demand scenario of identifying which song the current music corresponds to, audio retrieval technology has emerged. At present, audio retrieval technologies are mainly divided into two categories, one is audio retrieval technology based on audio fingerprints, and the other is based on fuzzy matching. Audio retrieval technology is mainly based on vector feature retrieval. Regardless of whether it is an audio retrieval technology based on audio fingerprints or an audio retrieval technology based on vector features, the target retrieval object is usually hundreds of thousands or more songs in the music library. The time-consuming and resource overhead of features is relatively large, and because there are a large number of interfering songs, it is very easy to mistakenly...

AI Technical Summary

Problems solved by technology

Regardless of whether it is an audio retrieval technology based on audio fingerprints or an audio retrieval technology based on vector features, the target retrieval object is usually hundreds of thousands or more songs in the music library. The time-consuming and resource overhead of features is relatively large, and because there are a large number of interfering songs, it is very easy to mistakenly recall similar features, which do not actually correspond to the correct target song. Therefore, in the scene of a star concert or TV party program, Difficult to quickly and accurately return a correct result based on existing audio retrieval techniques

To sum up, in the process of implementing the present invention, the inventor at least found that there are problems in the prior art that audio retrieval is time-consuming, resource-intensive, and not accurate

Images