Skip to content

Data Handling

Data Structure

The project works with audio data stored in a specific directory structure:

  • Raw Data: Unprocessed audio files, typically in .wav format.
  • data/raw/tracks/: Contains folders for each track (e.g., song1/, song2/).

  • Example: data/raw/tracks/song1/vocals.wav

  • Processed Data: Feature files and preprocessed data ready for model input.

  • data/processed/features/: Contains extracted features like MFCCs and spectrograms.
  • Example: data/processed/features/song1_vocals_mfcc.npy

Data Collection

  • Source Audio: Collect raw audio files for different tracks (vocals, instruments, etc.).
  • Reference Tracks: Include mixed and mastered tracks for the AI to learn from.

Data Processing Workflow

  1. Loading Audio:

  2. Use the load_audio() function from src/data_processing.py to load audio files into numpy arrays.

  3. Splitting Tracks:

  4. Use the split_tracks() function to divide longer audio files into manageable segments.

  5. Feature Extraction:

  6. Use extract_mfcc() and extract_spectrogram() from src/feature_extraction.py to extract features for model training.

  7. Saving Processed Data:

  8. Save extracted features as .npy files in the data/processed/features/ directory.

Ensure your data is properly organized before starting model training.