projects | Jordie Shier

Designing Percussive Timbre Remappings: Negotiating Audio Representations and Evolving Parameter Spaces

Exploring audio representations for drum-to-synthesizer timbre remapping using a technical-practice methodology and an approach for timbre remapping with non-differentiable synths using a genetic algorithm and neural networks for low-latency inference.

torchdrum

An audio-driven 808 drum synthesiser built with PyTorch and JUCE. ADC 2024 virtual poster outlining timbre remapping with DDSP and plug-in implementation.

Real-time Timbre Remapping with Differentiable DSP

Mapping timbre variations from an acoustic percussion instrument to a synthesizer in real-time with a differentiable drum synthesizer and a learned neural network for real-time performance.

BRAVE

Low-latency neural audio synthesis timbre transfer model based on the RAVE architecture.

Differentiable Modeling of Percussive Audio with Transient and Spectral Synthesis

Improving transient response for percussive audio synthesis using a combination of sinusoidal modelling techniques and neural audio synthesis.

Holistic Evaluation of Audio Representations

HEAR is an audio representation benchmark consisting of tasks across multiple domains. It was launched in 2021 as a NeurIPS challenge and includes an open-source evaluation toolkit, nineteen evaluation tasks, and an open leaderboard.

torchsynth

torchsynth is based upon traditional modular synthesis written in pytorch. It is GPU-optional and differentiable. Most synthesizers are fast in terms of latency. torchsynth is fast in terms of throughput. It synthesizes audio 16200x faster than realtime (714MHz) on a single GPU. This is of particular interest to audio ML researchers seeking large training corpora.