Sonora uses the official Moonshine WASM release version 0.1.5.
The Moonshine implementation is MIT, except separately licensed components.
This notice grants no license for Sonora's own application code.
Project and license: https://github.com/moonshine-ai/moonshine/tree/v0.1.5
Release archive: https://github.com/moonshine-ai/moonshine/releases/download/v0.1.5/moonshine-voice-wasm.tar.gz
SHA256: c515bf7691e12048f70a92cc82b3b0894c16c3773ffcacb7d48944fb150e4837
WASM SHA256: 22fca4a5b2dc50fe36dc68e4d25fab73b0250b71f744d9fea511c3a308b2420a
The release receipt is distributed in engine/release.json.

Moonshine streaming EN and DE model packs: MIT, according to the official catalog.
https://github.com/moonshine-ai/moonshine/blob/v0.1.5/docs/models/available-models.md

PYANNOTE COMMUNITY-1 PRETRAINED DATA - CC BY 4.0
Sonora uses pyannote Community-1 speaker diarization models and PLDA/xvector
parameters by pyannote.ai, licensed under Creative Commons Attribution 4.0
International. The neural models were converted and quantized to ONNX Runtime
format by Moonshine AI; Sonora uses those converted assets unchanged.
Original material: https://huggingface.co/pyannote/speaker-diarization-community-1
Conversion: https://github.com/moonshine-ai/moonshine/blob/v0.1.5/core/cpp-annote/README.md
License: https://creativecommons.org/licenses/by/4.0/
Full terms: licenses/CC-BY-4.0.txt
This attribution covers segmentation.ort, embedding.ort, and compiled
pretrained PLDA/xvector/config data. MIT wrapper headers do not remove the
original data's attribution requirements. No endorsement is implied.
Proprietary app terms must preserve recipients' CC BY rights in this material.

CPP-ANNOTE IMPLEMENTATION - MIT
Copyright (c) 2026 Useful Sensors, Inc. (dba Moonshine AI).
Full terms: licenses/cpp-annote-MIT.txt.

EIGEN - MPL 2.0
The upstream runtime builds cpp-annote with EIGEN_MPL2_ONLY=1.
Full terms: licenses/Eigen-MPL-2.0.txt.
The corresponding vendored Eigen source, including its per-file copyright
notices, is available under MPL 2.0 at:
https://github.com/moonshine-ai/moonshine/tree/234f60faa0eb388b01cdf7e60aca232af37aefda/core/third-party/Eigen
This source notice covers the Eigen files, not separately authored Sonora
application files. Sonora makes no Eigen changes. If rebuilding/modifying
the runtime, provide corresponding covered source for that build and preserve
recipients' MPL rights.

OTHER IMPLEMENTATION COMPONENTS
ONNX Runtime: MIT - licenses/ONNX-Runtime-MIT.txt.
Upstream aggregate: licenses/ONNX-Runtime-1.23.2-ThirdPartyNotices.txt.
That aggregate also lists optional components; it is not an exact linked SBOM.
Silero VAD: MIT, Copyright (c) 2020-present Silero Team - licenses/Silero-VAD-MIT.txt.
kaldi-native-fbank: Apache 2.0 - licenses/kaldi-native-fbank-Apache-2.0.txt.
KISS FFT: BSD 3-Clause, Copyright (c) 2003-2010 Mark Borgerding -
licenses/kissfft-COPYING.txt and licenses/kissfft-BSD-3-Clause.txt.
nlohmann/json: MIT - licenses/nlohmann-json-MIT.txt.
utf8cpp: Boost 1.0 - licenses/utf8cpp-Boost-1.0.txt.
utf8proc: MIT plus Unicode data notices - licenses/utf8proc-MIT-Unicode.txt.
Emscripten: MIT or Illinois/NCSA - licenses/Emscripten-4.0.8-LICENSE.txt.
Source details: licenses/README.md.
Further upstream component sources:
https://github.com/moonshine-ai/moonshine/tree/v0.1.5/core/third-party

Models are downloaded from their upstream CDN during explicit preparation;
they are not included in the application source repository.

WebM recording duration repair: fix-webm-duration 1.0.6, MIT.
Copyright (c) 2018 Yury Sitnikov. Full license: WEBM-LICENSE.txt.
https://github.com/yusitnikov/fix-webm-duration

WLLAMA (llama.cpp for WebAssembly) - MIT
Sonora uses @wllama/wllama 3.8.1, a WebAssembly build of llama.cpp with a JavaScript binding,
to run the optional language model on the CPU inside the browser. MIT licensed.
Project: https://github.com/ngxson/wllama   llama.cpp: https://github.com/ggml-org/llama.cpp (MIT)
Full terms: WLLAMA-LICENSE.txt. Served from llm/index.js and llm/wllama.wasm.

QWEN3 LANGUAGE MODELS - APACHE 2.0
Optional language model packs for summaries and dictation clean-up:
Qwen3-0.6B-Q4_K_M.gguf (396,705,472 bytes, SHA-256 ac2d97712095a558e31573f62f466a3f9d93990898b0ec79d7c974c1780d524a)
Qwen3-1.7B-Q4_K_M.gguf (1,107,409,472 bytes, SHA-256 b139949c5bd74937ad8ed8c8cf3d9ffb1e99c866c823204dc42c0d91fa181897)
Copyright Alibaba Cloud (Qwen team), Apache License 2.0. GGUF conversions published by Unsloth.
https://huggingface.co/Qwen/Qwen3-0.6B  https://huggingface.co/Qwen/Qwen3-1.7B
https://huggingface.co/unsloth/Qwen3-0.6B-GGUF  https://huggingface.co/unsloth/Qwen3-1.7B-GGUF
The models are downloaded by the user on demand and are not part of this distribution unless bundled by an administrator.

PARAKEET.JS - MIT
Sonora uses parakeet.js 1.4.4 (Copyright (c) ysdede, MIT) to run NVIDIA Parakeet speech models in the
browser. Full terms: PARAKEET-JS-LICENSE.txt. Served from hq/parakeet/ with the bare onnxruntime-web
import pointed at the local copy; nothing is fetched from a CDN.
https://github.com/ysdede/parakeet.js

ONNX RUNTIME WEB - MIT
Copyright (c) Microsoft Corporation. Licensed under the MIT License:
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.
onnxruntime-web 1.24.1, served from hq/ort/ (ort.wasm.min.mjs, ort-wasm-simd-threaded.mjs, .wasm).
https://github.com/microsoft/onnxruntime

NVIDIA PARAKEET TDT 0.6B V3 - CC-BY-4.0
The optional high-quality speech pack is NVIDIA's parakeet-tdt-0.6b-v3 (Creative Commons Attribution 4.0),
exported to ONNX by istupakov and quantised (int4 encoder, int8 decoder) by efederici:
encoder-model.int4.onnx (390,929,172 bytes, SHA-256 df4f1e5ff7a3af4e9d4b7078055164b11005e5d8a4c100e67f583c23975f7a31)
decoder_joint-model.int8.onnx (18,202,004 bytes, SHA-256 eea7483ee3d1a30375daedc8ed83e3960c91b098812127a0d99d1c8977667a70)
vocab.txt (93,939 bytes, SHA-256 d58544679ea4bc6ac563d1f545eb7d474bd6cfa467f0a6e2c1dc1c7d37e3c35d)
Attribution: NVIDIA Corporation, https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
ONNX exports: https://huggingface.co/istupakov/parakeet-tdt-0.6b-v3-onnx  https://huggingface.co/efederici/parakeet-tdt-0.6b-v3-onnx-int4
The model is downloaded by the user on demand and is not part of this distribution unless bundled by an administrator.