...IN PROGRESS...

The Beatles - Revolver (1966)

Revolver is the seventh studio album by the English rock band the Beatles.

You can listen to it on (link):

Data

• Spotify Features
  • Extracted with spotipy package
• Audio Wave
  • 14 songs in .wav files
  • Not uploaded in this repo due to a copyright issue
• Annotations Data (source)
  • Structures in csv
  • Keys in csv
• Other Metadata (source)
  • Lyrics
  • Lead vocals, writers

What I did with this album:

• Source Separation

  • Separate music into four sources: vocal, bass, drum, and other
  • Use pre-trained Hybrid Transformer Demucs model (using deep neural networks)

• Audio Feature Extraction

  • Extract audio features such as mean, variance, minimum, and maximum values of spectrogram features, mfccs, zcr, rmse, chromagram features, and tempo features
  • Use ftrosa package which is based on librosa

• Melody Extraction and Transcription: extracting a main melody of each song

  • Get F0(fundamental frequency) of main vocal melody
  • Piano transcription of the melodies. (in progress)

• Lyrics Parts

  • Create lyric wordcloud with sentiment analysis
  • Use wordcloud and nltk

---

TO-DOs

• Harmony Parts

  • Singer Identification: identifying a lead vocal (John, Paul, George, or Ringo) of each song
    • Use unsupervised learning to distinguish the vocals. Is it possible with spectral features?
    • Or you can train a ML/DL model on the other Beatles vocal data and test to identify the vocals.
  • Chord recognition: detecting chords, keys of each song

• Rhythm Parts

  • Drum Pattern Anaylsis
  • Beat tracking
    • Get BPM, detect onset
    • Drum Transcription

• Structural Parts

  • Structure Analysis and Segmentation

• ... and I'll come up with more ideas.