# Third-party software and model notices

Olenji is an independently authored application. It uses public projects and public model artifacts through adapters; it does not incorporate FluidVoice's desktop UI or private enhancement service.

The application uses FastAPI, Starlette, Pydantic, Uvicorn, websockets, httpx, NumPy, SoundFile, sounddevice, huggingface-hub, Rich and jsonschema. Optional inference environments use faster-whisper/CTranslate2, transcribe.cpp, sherpa-onnx, ONNX Runtime, SOXR, or the public FluidAudio Swift package. Windows installations include SoundCard for the loopback adapter. Native audio, shortcut and text-insertion helpers use operating-system frameworks; Linux text insertion uses the desktop’s AT-SPI accessibility service; Linux Wayland shortcuts use the desktop's consent-mediated GlobalShortcuts portal through Gio/GLib, when available. FFmpeg is a separate prerequisite and its applicable license depends on the installed build. jsdom and Playwright are development-only browser-test dependencies; browsers and their automation runtimes are not bundled in the application wheel.

These dependencies are obtained from their normal package registries or pinned upstream sources. Their licenses are not replaced by Olenji's Apache-2.0 license. The release source and wheel do not bundle model weights, native inference binaries, FFmpeg, Python itself, or an entire optional runtime environment.

Model metadata records publisher URLs, artifact identities, expected license information, platform restrictions, and exact-versus-alternative parity provenance. `olenji models inspect ID` and `olenji models plan-install` expose this information before download. Installation manifests retain immutable source revisions and verified file digests where available.

Publisher conversions may use a different license from an upstream model. For example, the audited Cohere Core ML conversion declares GPL-3.0. Fluid Speech Pico is available through its public publisher-pinned archive, whose bytes and hashes were verified during this audit. The archive did not include a separate weight license, so the catalog reports `unresolved` and does not grant redistribution rights or bundle those weights. Apple assets are installed and managed through supported operating-system mechanisms rather than repackaged into Olenji.

The public speech fixture used for real inference verification is the `tests/jfk.flac` recording in OpenAI's Whisper repository: https://github.com/openai/whisper/blob/main/tests/jfk.flac. It is used only in the validation workspace; it is not bundled as user content or a fake transcript in the production application.

The bundled browser uses locally authored CSS, vanilla JavaScript and inline SVG symbols. It does not load remote fonts, analytics, icon CDNs or advertising scripts.

## Portable speaker adapters (1.2.0)

The additional speaker adapters are independently implemented from public model/exporter contracts. They do not incorporate OpenVoiceEverything's GPL application source or relicense its Core ML packages. Runtime/source license and model-weight license are separate.

- **parakeet-rs**, by Enes Altun, supplies the acknowledged primary Nemotron/Sortformer tensor/cache and frontend reference. Reviewed source revisions are `50ef467dbaefdbff2d9fd81e3c1bc30358c6f278` (Nemotron) and `cd5888be5e99724299d5ffb2af744a150c551dbd` (earlier Sortformer v2 support). MIT notice reproduced below. [Source](https://github.com/altunenes/parakeet-rs), [license at the reviewed revision](https://github.com/altunenes/parakeet-rs/blob/50ef467dbaefdbff2d9fd81e3c1bc30358c6f278/LICENSE).
- **LS-EEND ONNX** exports/reference runtime by `GradientDescent2718`, revision `cc40a1e1242c148fbbc15c132e43b8ac15056e53`, publishes four dataset-specific model variants under MIT. They originate from LS-EEND by Di Liang and Xiaofei Li. The acoustic feature reference identifies Copyright 2019 Hitachi, Ltd. (author: Yusuke Fujita), under MIT; its notice is retained below. Dataset access and usage terms for AMI, CALLHOME and DIHARD remain with their providers. [Publisher and reference files](https://huggingface.co/GradientDescent2718/LS-EEND-ONNX/tree/cc40a1e1242c148fbbc15c132e43b8ac15056e53), [original research source](https://github.com/Audio-WestlakeU/FS-EEND).
- **Pyannote segmentation 3.0** is published under MIT (Copyright 2022 CNRS); its ONNX export is supplied through sherpa-onnx. The paired **WeSpeaker** pretrained weights retain their corresponding dataset terms, including CC BY 4.0 for VoxCeleb models. The Apache-2.0 license of runtime/code does not automatically license pretrained weights. This adapter is not the gated pyannote community-1 pipeline. [Segmentation export and license](https://huggingface.co/csukuangfj/sherpa-onnx-pyannote-segmentation-3-0/tree/9403a6902bb58e3d5ae8c7e77c3422de279db2e0), [WeSpeaker pretrained-model terms](https://github.com/wenet-e2e/wespeaker/blob/master/docs/pretrained.md).
- **NVIDIA Nemotron 3 Diarization** weights use [OpenMDW 1.1](https://openmdw.ai/license/1-1/). **Sortformer v2** weights use CC BY 4.0, while **Sortformer v2.1** uses the [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/). Their separately identified ONNX conversions retain the weight terms; a converter's MIT code license does not replace those terms.

These weights, datasets, inference binaries and optional environments are not bundled into Olenji. Explicit model installation fetches the reviewed data files and optional dependencies from their publishers; it does not execute downloaded repository Python/Rust scripts. Catalog inclusion is not a redistribution permission, accuracy comparison or claim of numerical equivalence with OVE.

### MIT notices

Acknowledged primary components:

- parakeet-rs: Copyright (c) 2025 Enes Altun.
- LS-EEND acoustic feature reference: Copyright 2019 Hitachi, Ltd. (author: Yusuke Fujita).
- Pyannote segmentation export: Copyright (c) 2022 CNRS.

The following MIT terms accompany those upstream notices:

> Permission is hereby granted, free of charge, to any person obtaining a copy
> of this software and associated documentation files (the "Software"), to deal
> in the Software without restriction, including without limitation the rights
> to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
> copies of the Software, and to permit persons to whom the Software is
> furnished to do so, subject to the following conditions:
>
> The above copyright notice and this permission notice shall be included in all
> copies or substantial portions of the Software.
>
> THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
> AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
> OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
> SOFTWARE.