Mad Mashup Generator (with madmom beats)
The mashup consists of an original and a cover version of your favourite song. Their chorus segments are played in alternating way bar-by-bar based on the music bars. Hack done during https://www.ismir2020.net/hamr/ following the https://www.ismir2020.net/ conference.
License: MIT LICENSE
Requirements
pip install -r requirements.txt
Documentation
The hack is based on state of the art MIR methods for beat tracking, downbeat tracking, music key detection and lyric-to-audio alignment.
Step 1
Lyrics-to-Audio Alignment based on the MIREX 2020 winning method [1] and take the chorus audio segment based on the tagged "chorus" section in Genius.com
Step 2
Detect tempo based on detected beats [2] implemented in the https://github.com/CPJKU/madmom.
Time stretching using the librosa https://librosa.org/ in order to match the tempo
Step 3
Key detection based on [3] implemented in https://github.com/CPJKU/madmom and then pitch-shifting with https://pypi.org/project/rubberband/
Step 4 Mashup:
Detect downbeats (music bars) with downbeat tracking [2] as implemented in https://github.com/CPJKU/madmom.
REFERENCES:
[1] Gao et al. - Lyrics Transcription And Lyrics-to-audio Alignment With Music-informed Acoustic Models, MIREX 2020 https://www.music-ir.org/mirex/abstracts/2020/GL1.pdf
[2] Böck et al. - Joint Beat and Downbeat Tracking with Recurrent Neural Networks, Proceedings of the 17th International Society for Music Information Retrieval Conference (ISMIR), 2016.
[3] Korzeniowski et al. - Genre-Agnostic Key Classification with Convolutional Neural Networks, Proceedings of the 19th International Society for Music Information Retrieval Conference (ISMIR), 2018.