Abstract
Music is frequently modified from an original version before uploading to music-sharing or video-sharing websites or social media networks. For example, the music can be remixed, have voice-overs added, or have other edits. In several use cases, e.g., search, deduplication, ensuring fair use, etc. it is of interest to determine if an uploaded audio track is substantially similar to existing audio tracks in a database. However, modifications made to the original version can in some instances be enough that a match is not obtained with any track in the reference database even when the tracks match substantially. This disclosure describes neural network based techniques to ignore modifications, e.g., voice-overs, from an audio track such that a match, if any, with a reference audio track is easier to detect.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Pasula, Hanna Maria and Seiler, Martin, "Boosted Audio Recall for Music Matching", Technical Disclosure Commons, (June 29, 2020)
https://www.tdcommons.org/dpubs_series/3369