SongSpy is a web application that utilizes advanced audio extraction techniques to analyze the timbre and other acoustic features of a song. It employs machine learning techniques to train and identify patterns specific to different artists. By examining characteristics such as tonal quality, instrument sounds, and rhythmic patterns, the model can accurately classify the artist behind a given piece of music. It also includes a feature to determine if the song is artificially generated by comparing it to the artist's current music catalogue.
The use of the deep learning model ‘Diff-SVC’ to transform voices of producers and everyday artists into those of their favorite artists has led to a surge in copyright claims. Platforms like YouTube and Soundcloud are now flooded with AI-generated remakes and covers, prompting labels and streaming services to evaluate how to address this new technology.
Regardless of an artist’s stance on the new technology, it is necessary to implement a flagging system that notifies streaming services when artificially generated tracks utilizing an artist’s voice are uploaded. This would benefit streaming platforms by automatically identifying illegal content and help labels save legal and administrative resources in pursuing copyright claims.
Final Product --> SongSpy
The current version of the model is trained on the following artists:
- Drake
- Post Malone
- The Weeknd
- SZA
- Kanye West
- Taylor Swift
- Lil Uzi Vert
- Juice WRLD
- Travis Scott
- Upload a song in MP3/WAV/AIF format to the web application.
- The model will analyze the song's timbre and other acoustic features.
- The application will classify the artist behind the song and determine if it is artificially generated or not.
SongSpy is released under the MIT License.
If you have any questions or suggestions, please feel free to contact me at smullins998@gmail.com