This project is a submission for the Ai Music Generation Challenge 2022, specifically focusing on the sub-challenge of generating plausible reels using RAVE (Realtime Audio Variational autoEncoder).
This project aims to build an artificial system that generates reels, adhering to the standards and characteristics of the reels published in F. O'Neill's "The Dance Music of Ireland: O'Neill's 1001" (1907). RAVE, a Realtime Audio Variational autoEncoder, is utilized for fast and high-quality audio waveform synthesis.
The system consists of the following components:
- Data preprocessing: Preparing the training data by extracting and transforming the reels from F. O'Neill's collection into a suitable format for training RAVE.
- Model architecture: Implementing the RAVE model architecture for audio waveform synthesis.
- Training procedure: Training RAVE using the preprocessed reel dataset, incorporating the two-stage training procedure (representation learning and adversarial fine-tuning) mentioned in the RAVE paper.
- Reel generation: Utilizing the trained RAVE model to generate a collection of plausible reels.