mir

Music Information Retrieval

Setup

Command line (on Linux):

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txtpip

Use Jupyter notebooks (like this example from FMP website.)
Analyse keyboard intro to Four out of Five:
- Time 4-16s
- Apply chord recognition from FMP p256.
- Infer tuning (centre frequency) and align chroma on that.
- Extract melody: see FMP chapter 8 and https://www.audiolabs-erlangen.de/resources/MIR/FMP/C8/C8S2_MelodyExtractSep.html
- Nonnegative matrix factorisation (NMF) -based audio decomposition: template matrix of MIDI note numbers and their frequencies (fundamental and harmonics) and activation matrix (note numbers vs time).
- NMF for a track with known score (C major scale piano?), use to tune parameters to reconstruct it then apply to tracks with unknown score.
- Apply to Salieri's march of welcome.
Save plots to file.
Analyse songs in data folder:
- Chromagram
- Audio thumbnail
- Chord detection
- Self-similarity matrix
- Tempo analysis
- Separate harmonic and percussive, save to files, listen (and adjust params if need be).
Add songs to data folder: Beatles, White Stripes.
Get PySoundFile working for load?
Use Viterbi to truncate silence at beginning and end (only).
Use Laplacian segmentation to identify and label song sections (verse, chorus etc.).
Play audio with vertical line sweeping through plot (e.g. chroma) in sync. See FMP toolbox in YouTube demo.