GigFrame is built on the work of many open-source authors and research groups. This page lists the AI models, audio engines and libraries we use, with version, license, source and authors. Where we could not verify something, it says NOT VERIFIED instead of guessing.
These run on our own servers to analyse the audio and documents you give us.
| Name | What it does in GigFrame | Version | License | Source | Authors |
|---|---|---|---|---|---|
| ISMIR2019 Large-Vocabulary Chord Recognition | Chord recognition | Repository HEAD at image build (not pinned) | MIT | github.com/music-x-lab/ISMIR2019-Large-Vocabulary-Chord-Recognition | Junyan Jiang, Ke Chen, Wei Li, Gus Xia (Music X Lab), © Music X Lab |
| faster-whisper | Speech-to-text runtime (lyrics) | 1.2.1 | MIT | github.com/SYSTRAN/faster-whisper | SYSTRAN |
| Whisper large-v3-turbo (CTranslate2 conversion) | Speech-to-text model (lyrics). The GPU profile uses Whisper large-v3 (also MIT). | large-v3-turbo | MIT | huggingface.co/openai/whisper-large-v3-turbo | OpenAI (conversion: Mobius Labs) |
| WhisperX | Installed in the lyrics runner (word alignment is currently not used) | 3.8.6 | BSD-2-Clause | github.com/m-bain/whisperX | Max Bain |
| basic-pitch | Note and tab transcription | 0.4.0 | Apache-2.0 | github.com/spotify/basic-pitch | Spotify |
| torchcrepe (tiny model) | Pitch tracking for tab | Not pinned | MIT | github.com/maxrmorrison/torchcrepe | Max Morrison |
| Demucs (htdemucs / htdemucs_6s) | Splitting audio into parts (stems) | 4.0.1 | Code: MIT. Pretrained weights: released by the authors for research purposes; commercial terms not confirmed. | github.com/facebookresearch/demucs | Alexandre Défossez / Meta AI |
About Demucs: the code is MIT, but the pretrained weights were released for research purposes and we have not confirmed commercial terms. Per our decision D-306, splitting for paying users moves to a commercially licensed service. Until that switch is complete, splitting runs through fal.ai (which hosts Demucs), then Replicate as a second option, and as a last fallback Demucs on our own server.
Supporting libraries in the same runners:
| Name | Version | License | Source |
|---|---|---|---|
| PyTorch | 2.8.0 / 2.4.1 | BSD-3-Clause | github.com/pytorch/pytorch |
| onnxruntime | 1.30.0 | MIT | github.com/microsoft/onnxruntime |
| librosa | Not pinned | ISC | github.com/librosa/librosa |
| CTranslate2 | 4.8.2 | MIT | github.com/OpenNMT/CTranslate2 |
| mir_eval | Not pinned | MIT | github.com/mir-evaluation/mir_eval |
| pretty_midi | Not pinned | MIT | github.com/craffel/pretty-midi |
We tested these but they do not run for you. They are listed so nobody has to wonder.
| Name | What it does in GigFrame | Version | License | Source | Authors |
|---|---|---|---|---|---|
| CREMA | Evaluated as a chord engine; not in the production image | 0.2.0 | BSD-2-Clause (GitHub) / ISC (PyPI 0.2.0) — NOT VERIFIED which one applies to 0.2.0 | github.com/bmcfee/crema | Brian McFee (with Juan P. Bello) |
| madmom | Internal evaluation only; never shipped | 0.16.1 | Code: BSD. Model files: CC BY-NC-SA (non-commercial) | github.com/CPJKU/madmom | Sebastian Böck et al. (CP JKU) |
| ACE-Step | Music generation; switched off (decision D-302) | 1.5 | MIT | github.com/ace-step/ACE-Step-1.5 | ACE Studio / StepFun |
These are services we call, not software we ship, so there is no license to list. Each is governed by the provider's own terms. What data goes where is explained in our privacy policy.
| Name | What it does in GigFrame | Terms |
|---|---|---|
| fal.ai | Stem separation (hosts Demucs) | fal.ai/terms |
| Replicate | Stem separation, second option (hosts Demucs) | replicate.com/terms |
| AssemblyAI (EU) | Speech-to-text for lyrics | assemblyai.com/legal/terms-of-service |
| OpenRouter → Google Gemini 3.1 Flash Lite | PDF and photo import, sheet-music reading, translations | openrouter.ai/terms |
| Unsplash | Cover images | unsplash.com/license |
The libraries bundled into the GigFrame Studio web app and the server. Versions as locked in our package files.
| Name | Version | License | Source |
|---|---|---|---|
| preact | 10.29.0 | MIT | github.com/preactjs/preact |
| preact-router | 4.1.2 | MIT | github.com/preactjs/preact-router |
| tone | 15.1.22 | MIT | github.com/Tonejs/Tone.js |
| @tonejs/midi | 2.0.28 | MIT | github.com/Tonejs/Midi |
| tonal (@tonaljs/chord-detect, @tonaljs/midi) | 4.9.1 / 4.10.2 | MIT | github.com/tonaljs/tonal |
| soundfont-player | 0.12.0 | MIT | github.com/danigb/soundfont-player |
| @coderline/alphatab (alphaTab) | 1.8.4 | MPL-2.0 | github.com/CoderLine/alphaTab |
| opensheetmusicdisplay | 2.0.0 | BSD-3-Clause | github.com/opensheetmusicdisplay/opensheetmusicdisplay |
| pdfjs-dist (PDF.js) | 6.2.108 | Apache-2.0 | github.com/mozilla/pdf.js |
| jspdf | 4.2.1 | MIT | github.com/parallax/jsPDF |
| html2canvas | 1.4.1 | MIT | github.com/niklasvh/html2canvas |
| jszip | 3.10.1 | Dual-licensed: MIT or GPL-3.0-or-later (we use it under MIT) | github.com/Stuk/jszip |
| @breezystack/lamejs | 1.2.7 | LGPL-3.0. The source is available at the link on the right. | github.com/shijinyu/lamejs |
| CodeMirror 6 (@codemirror/*) | 6.x (state 6.6.0, view 6.40.0) | MIT | github.com/codemirror |
| qrcode | 1.5.4 | MIT | github.com/soldair/node-qrcode |
| sortablejs | 1.15.7 | MIT | github.com/SortableJS/Sortable |
| sql.js | 1.14.1 | MIT | github.com/sql-js/sql.js |
| nosleep.js | 0.12.0 | MIT | github.com/richtr/NoSleep.js |
| hono | 4.13.3 | MIT | github.com/honojs/hono |
| better-auth | 1.6.30 | MIT | github.com/better-auth/better-auth |
| zod | 4.4.3 | MIT | github.com/colinhacks/zod |
| drizzle-orm | 0.45.2 | Apache-2.0 | github.com/drizzle-team/drizzle-orm |
The instrument samples we load for playback come from the MusyngKite soundfont in the midi-js-soundfonts collection by Benjamin Gleitzman, licensed CC BY-SA 3.0.
| Name | Version | License | Source |
|---|---|---|---|
| midi-js-soundfonts (MusyngKite) | Not pinned | CC BY-SA 3.0 | github.com/gleitz/midi-js-soundfonts |
Thank you to every author and research group above. GigFrame would not exist without your work. Spotted a mistake or a missing credit? Email [email protected] and we will fix it.
Last updated: 2026-10-09