deepraaga-preprocess

Data ingestion layer for DeepRaaga. Extracted from the original DeepRaaga project.

Installation

pip install deepraaga-preprocess

Overview

The deepraaga-preprocess module handles the ingestion and transformation of Carnatic music data. This includes parsing MIDI files and abstract sequence processing to map musical notes and swaras into sequences readable by machine learning models.

Usage

You can use the DataProcessor to easily convert a directory of MIDI files into numpy feature sequences ready for training:

import os
from deepraaga_preprocess.data_processor import DataProcessor

processor = DataProcessor(sequence_length=100)

# Process a directory of raw MIDI files and output training numpy arrays
processor.process_dataset(midi_dir='data/raw_midi', output_dir='data/processed')

# Load the resulting vocabulary mapping later
processor.load_vocab('data/processed/vocab.pkl')

Features

• MIDI Feature Extraction: Parses incoming MIDI structures using music21 and resolves them to sequential sequences.
• Dynamic Vocabulary: Dynamically builds note-to-integer mappings.
• Raga Abstraction: Support for processing Carnatic arohanam and avarohanam notation (via preprocess_raga.py).

License

This project is licensed under the MIT License.