Reproduce “Towards zero-shot amplifier modeling: One-to-many amplifier modeling via tone embedding control” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yu-Hua Chen, Yen-Tung Yeh, Yuan-Chiao Cheng, Jui-Te Wu, Yu-Hsiang Ho, Jyh-Shing Roger Jang, Yi-Hsuan Yang
The research from National Taiwan University and Positive Grid introduces a single neural model for "one-to-many" guitar amplifier emulation, leveraging a contrastive learning-based tone embedding encoder and a FiLM-conditioned GCN generator for "zero-shot" tone transfer. The model successfully generalizes to unseen amplifier tones, exhibiting complex STFT losses that are 1-2 times larger for unseen low-gain amps and approximately 2 times larger for unseen high-gain amps compared to seen counterparts.
Key insights
- Contrastive learning successfully trains an encoder to disentangle tone (style) from musical content, yielding robust tone embeddings that accurately cluster similar amplifier tones.
- FiLM serves as an effective conditioning method for injecting tone embeddings into a neural audio effect model, outperforming simpler concatenation approaches.
- Training the tone embedding encoder with "unpaired reference" audio (same tone, different musical content) can enhance the robustness of tone representations and model performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.