# Third-party software and models

Noteform runs in your browser. Nothing below sends audio anywhere. Full license texts are served with the site
in `licenses/` (source: `frontend/public/licenses/`); the licenses of all bundled npm packages are generated at
build time into `licenses/javascript-packages.txt`.

## Shipped with the website

**Basic Pitch** by Spotify (Rachel M. Bittner, Juan José Bosch, David Rubinstein, Gabriel Meseguer-Brocal,
Sebastian Ewert), version 0.4.0, Apache License 2.0, https://github.com/spotify/basic-pitch. Copyright 2022
Spotify AB. Noteform redistributes the pretrained ICASSP 2022 model in its ONNX export, unmodified, from the
`basic-pitch` 0.4.0 package (`saved_models/icassp_2022/nmp.onnx`, SHA-256
`2c3c1d144bfa61ad236e92e169c13535c880469a12a047d4e73451f2c059a0ec`), as
`frontend/src/engine/model/basic-pitch-icassp2022-nmp.onnx`. `frontend/src/engine/basicPitch.ts` is a TypeScript
port of Basic Pitch's `inference.py` and `note_creation.py`. License and NOTICE: `basic-pitch-LICENSE.txt`,
`basic-pitch-NOTICE.txt`. If you use Basic Pitch results in research, cite:

> Bittner, R. M., Bosch, J. J., Rubinstein, D., Meseguer-Brocal, G., & Ewert, S. (2022). A Lightweight
> Instrument-Agnostic Model for Polyphonic Note Transcription and Multipitch Estimation. ICASSP 2022.

**ONNX Runtime Web** (`onnxruntime-web` 1.30.0) by Microsoft, MIT License. Its JavaScript and WebAssembly files
run the model. License and third-party notices: `onnxruntime-LICENSE.txt`, `onnxruntime-ThirdPartyNotices.txt`.

**librosa** 0.11.0, ISC License, Copyright (c) 2013--2023, librosa development team.
`frontend/src/engine/pyin.ts` is a TypeScript port of librosa's `pyin`, `onset_detect`, `peak_pick`, `rms` and mel
filter code (`librosa-LICENSE.txt`). pYIN: Mauch, M. & Dixon, S. (2014), pYIN: A fundamental frequency
estimator using probabilistic threshold distributions. ICASSP 2014.

**React, React DOM, scheduler, zod** (MIT): see `licenses/javascript-packages.txt`.

## Development only (not shipped)

Vite, TypeScript, ESLint, Prettier, Vitest, Playwright, Wrangler and `@cloudflare/vite-plugin` (MIT /
Apache-2.0), pinned in `frontend/package-lock.json`.

The Python reference (`reference/`, pinned in `reference/uv.lock`): basic-pitch 0.4.0 (Apache-2.0), onnxruntime
(MIT), librosa and resampy (ISC), numpy, scipy, scikit-learn, numba, soundfile (BSD), pydantic, mir_eval,
pretty_midi (MIT).

## Audio fixtures and demo

All test fixtures (`reference/tests/fixtures/audio`) and the demo recording are synthesized by code in this
repository (`reference/tests/fixtures/make_fixtures.py`, `frontend/src/engine/demo.ts`): additive tones and a
Karplus–Strong plucked-string model. No third-party recordings are included.