Attribution

Original work

The sound texture synthesis algorithm implemented here is by:

Jeffrey N. Chadwick (Cornell University, at the time of publication)
Doug L. James (Cornell University, at the time of publication; now Stanford University)

Published in:

Jeffrey N. Chadwick and Doug L. James, “Animating Fire with Sound,” ACM Transactions on Graphics (SIGGRAPH 2011), 30(4), August 2011. Section 5 (“Synchronized Sound Texture Synthesis”).

Project page (canonical reference): https://www.cs.cornell.edu/projects/Sound/fire/

Paper PDF: https://www.cs.cornell.edu/projects/Sound/fire/FireSound2011.pdf

Reference implementation

The Python and JavaScript code in this repository is a faithful port of the C++ reference implementation released by the authors at the URL above (sound_texture_synthesis.zip / sound_texture_synthesis.tar.gz). All algorithm design choices, parameter defaults, and numerical details follow that implementation. Differences from the released C++ are documented in docs/ALGORITHM.md.

The original C++ release used the ANN approximate-nearest-neighbour library. This port uses exact nearest-neighbour search (scipy.spatial.cKDTree in Python; a hand-port of the bundled KDTreeMulti template in JavaScript) so the two ports produce deterministic, bit-comparable output.

License

The original C++ implementation is distributed under the BSD 2-Clause License (see LICENSE). All derivative source files in this repository carry forward the original copyright notice and disclaimer.

Additional code (the Python and JavaScript ports, the browser demo, the test harness) is also licensed under the BSD 2-Clause License, with attribution to the original authors required by the same terms.

Training audio

The six WAV files under assets/training_audio/ are from The Recordist’s Ultimate Fire sound library (therecordist.com). The original Cornell release states:

“Permission to provide a few of these audio clips as examples in our source code package has been generously granted by The Recordist.”

This demo redistributes them as a derivative work of the same source-code package under the same permission. They are used unmodified (downmixed to 44.1 kHz mono in the original release). All copyright in the audio remains with The Recordist.

Bundled flame-simulation signals

The input.bin files for the five examples (burning_brick, candle, dragon, flame_jet, torch) are the low-frequency outputs of physically based flame simulations produced by the original authors and distributed with their code release. They are used here unmodified (re-encoded as Float32 for browser delivery). These are the same signals used by the sibling spectral bandwidth-extension demo at fire-bandwidth-extension/.

Browser demo context

This browser demo was developed for the course CS 448Z (Physically Based Animation and Sound) at Stanford University, Spring 2026, taught by Doug L. James. It is intended as an educational exploration of the algorithm and is not affiliated with or endorsed by the original publication beyond the use of its released code and data.

Vendored libraries

fft.js by Fedor Indutny (github.com/indutny/fft.js), MIT License. See fft.js at the project root for the original license header.