Core Concepts
This study presents a novel approach to generating high-quality classical Indian tabla music using advanced deep learning architectures, including Bi-LSTM with attention and transformer models.
Abstract
The paper explores music generation using deep learning techniques, with a focus on generating classical Indian tabla music.
Key highlights:
Extensive work has been done on generating piano and other Western music, but there is limited research on generating classical Indian music due to the scarcity of Indian music in machine-encoded formats.
The authors first experimented with LSTM-based models for generating classical piano music, achieving promising results. They then extended these techniques to generate tabla music.
For tabla music generation, the authors developed a novel Bi-LSTM with attention mechanism model and a transformer model, trained on a dataset of tabla waveform files.
The Bi-LSTM model achieved a loss of 4.042 and MAE of 1.0814, while the transformer model achieved a loss of 55.9278 and MAE of 3.5173 for tabla music generation.
The generated tabla music exhibits a harmonious fusion of novelty and familiarity, pushing the boundaries of music composition.
The authors discuss potential future work, such as enhancing the models by training on a larger tabla dataset, exploring generation for other classical Indian instruments, and generating multi-instrumental music fusing Indo-Western styles.
Stats
The Bi-LSTM model achieved a loss of 4.042 and MAE of 1.0814 for tabla music generation.
The transformer model achieved a loss of 55.9278 and MAE of 3.5173 for tabla music generation.
Quotes
"The resulting music embodies a harmonious fusion of novelty and familiarity, pushing the limits of music composition to new horizons."