THIRD-PARTY SOFTWARE LICENSES
=============================

DeepVideo / Echo Frame
Copyright (c) 2026 DeepForge Hub. All rights reserved.

This product includes software and resources developed by third parties.
Below is a list of third-party components, their licenses, and copyright holders.

================================================================================
1. FRAMEWORK & RUNTIME
================================================================================

--- Tauri (Desktop Application Framework) ---
License: MIT / Apache-2.0
Copyright (c) 2020-2024 Tauri Contributors
URL: https://tauri.app

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.

--- PyO3 (Rust-Python Bridge, release branch only) ---
License: MIT
Copyright (c) 2017-2024 PyO3 Contributors
URL: https://github.com/PyO3/pyo3

--- Python Runtime (release branch only) ---
License: PSF License (Python Software Foundation License)
Copyright (c) 2001-2024 Python Software Foundation
URL: https://www.python.org

================================================================================
2. PYTHON DEPENDENCIES (release branch only)
================================================================================

Note: These dependencies are used by the release_1.0.0 branch (Python backend).
The master branch uses pure Rust and does not require any of these.

--- faster-whisper (Speech Recognition) ---
License: MIT
Copyright (c) Guillaume Klein / SYSTRAN
URL: https://github.com/SYSTRAN/faster-whisper

--- onnxruntime (ONNX Model Inference) ---
License: MIT
Copyright (c) Microsoft Corporation
URL: https://github.com/microsoft/onnxruntime

--- onnx (Open Neural Network Exchange) ---
License: Apache-2.0
Copyright (c) ONNX Contributors
URL: https://github.com/onnx/onnx

--- numpy (Numerical Computing) ---
License: BSD-3-Clause
Copyright (c) 2005-2024 NumPy Developers
URL: https://numpy.org

--- scipy (Scientific Computing) ---
License: BSD-3-Clause
Copyright (c) 2001-2002 Enthought, Inc.; 2003-2024 SciPy Developers
URL: https://scipy.org

--- soundfile (Audio I/O) ---
License: BSD-3-Clause
Copyright (c) Bastian Bechtold
URL: https://github.com/bastibe/python-soundfile

--- srt (Subtitle Parser) ---
License: MIT
Copyright (c) Chris Down
URL: https://github.com/cdown/srt

--- sentencepiece (Tokenization) ---
License: Apache-2.0
Copyright (c) Taku Kudo / Google Inc.
URL: https://github.com/google/sentencepiece

--- huggingface-hub (Model Download) ---
License: Apache-2.0
Copyright (c) Hugging Face, Inc.
URL: https://github.com/huggingface/huggingface_hub

--- boto3 (AWS SDK / Cloudflare R2) ---
License: Apache-2.0
Copyright (c) Amazon Web Services
URL: https://github.com/boto/boto3

--- tqdm (Progress Bar) ---
License: MPL-2.0 / MIT
Copyright (c) Casper da Costa-Luis et al.
URL: https://github.com/tqdm/tqdm

--- requests (HTTP Library) ---
License: Apache-2.0
Copyright (c) Kenneth Reitz
URL: https://github.com/psf/requests

--- ctranslate2 (CTranslate2 Inference) ---
License: MIT
Copyright (c) OpenNMT Contributors
URL: https://github.com/OpenNMT/CTranslate2

--- av / PyAV (Multimedia) ---
License: BSD-3-Clause
Copyright (c) Mike Boers / PyAV Contributors
URL: https://github.com/PyAV-org/PyAV

--- tokenizers (Hugging Face Tokenizers) ---
License: Apache-2.0
Copyright (c) Hugging Face, Inc.
URL: https://github.com/huggingface/tokenizers

--- protobuf (Protocol Buffers) ---
License: BSD-3-Clause
Copyright (c) Google Inc.
URL: https://github.com/protocolbuffers/protobuf

--- flatbuffers ---
License: Apache-2.0
Copyright (c) Google Inc.
URL: https://github.com/google/flatbuffers

--- imageio-ffmpeg (FFmpeg Python Wrapper) ---
License: BSD-2-Clause
Copyright (c) imageio contributors / Almar Klein
URL: https://github.com/imageio/imageio-ffmpeg

--- demucs (Source Separation) ---
License: MIT
Copyright (c) Alexandre Défossez / Meta Platforms, Inc.
URL: https://github.com/adefossez/demucs

--- WeTextProcessing (Text Normalization) ---
License: Apache-2.0
Copyright (c) WeNet Contributors
URL: https://github.com/wenet-e2e/WeTextProcessing

--- WeTextProcessing FST Weight Files (Text Normalization Data) ---
License: Apache-2.0
Source: ModelScope pengzhendong/wetext
Original: WeTextProcessing (wenet-e2e, Apache-2.0)
Copyright (c) WeNet Contributors / Pengzhendong Guo
URL: https://modelscope.cn/models/pengzhendong/wetext

--- pyannote.audio (Speaker Diarization) ---
License: MIT
Copyright (c) pyannote contributors
URL: https://github.com/pyannote/pyannote-audio

--- sherpa-onnx (Speaker Diarization Backend) ---
License: Apache-2.0
Copyright (c) sherpa-onnx contributors
URL: https://github.com/k2-fsa/sherpa-onnx

================================================================================
3. MACHINE LEARNING MODELS
================================================================================

--- htdemucs / htdemucs-fp16 (ONNX) ---
License: MIT
Source: StemSplitio/htdemucs-onnx (Hugging Face)
Original: Meta Platforms, Inc. (MIT)
Copyright (c) Meta Platforms, Inc.

--- htdemucs-mlx (MLX, Apple Silicon GPU) ---
License: MIT
Source: mlx-community/demucs-mlx (Hugging Face)
Original: Meta Platforms, Inc. (MIT)
Copyright (c) MLX Community / Meta Platforms, Inc.

--- MOSS-TTS-Nano-100M-ONNX (FP32 & FP16) ---
License: Apache-2.0
Source: OpenMOSS-Team/MOSS-TTS-Nano-100M-ONNX (Hugging Face)
Copyright (c) OpenMOSS Team / Shanghai AI Lab

--- MOSS-TTS-Nano-100M-MLX ---
License: Apache-2.0
Source: mlx-community/MOSS-TTS-Nano-100M (Hugging Face)
Copyright (c) MLX Community / OpenMOSS Team

--- MOSS-Audio-Tokenizer-Nano-ONNX ---
License: Apache-2.0
Source: OpenMOSS-Team/MOSS-Audio-Tokenizer-Nano-ONNX (Hugging Face)
Copyright (c) OpenMOSS Team / Shanghai AI Lab

--- sherpa-onnx pyannote segmentation 3.0 (ONNX) ---
License: Apache-2.0
Source: csukuangfj/sherpa-onnx-pyannote-segmentation-3.0 (Hugging Face)
Original model: pyannote/segmentation-3.0 (MIT)
Copyright (c) sherpa-onnx contributors / pyannote contributors

--- 3D-Speaker ERes2Net ONNX (Speaker Embedding) ---
License: Apache-2.0
Source: protocolvoice/asr-models (Hugging Face; card tag license:apache-2.0, read 2026-09-25)
Original: ModelScope iic/speech_eres2net_large_sv_zh-cn_3dspeaker_16k
  (API License field "Apache License 2.0", read 2026-09-25; INTERSPEECH 2023, 3D-Speaker)
Copyright (c) Alibaba DAMO Academy / 3D-Speaker contributors
Verification (2026-09-25): the shipped speaker_embedding.onnx names its own origin in
  its trailing metadata_props - framework="3d-speaker",
  url="https://www.modelscope.cn/models/iic/speech_eres2net_large_sv_zh-cn_3dspeaker_16k/summary",
  comment="This model is from iic/speech_eres2net_large_sv_zh-cn_3dspeaker_16k",
  sample_rate=16000, output_dim=512, feature_normalize_type="global-mean".
  So the identity of the upstream checkpoint is established from the file itself, not
  from a converter's README. An earlier revision of this file credited "WeSpeaker",
  which is a different (also Apache-2.0) toolkit; the attribution is to 3D-Speaker.

--- Silero VAD (Voice Activity Detection) ---
License: MIT
Source: csukuangfj/vad (Hugging Face; the card declares no license tag)
Original: Silero Team (https://github.com/snakers4/silero-vad, MIT per the repository
  license metadata, read 2026-09-25)
Copyright (c) Silero Team / sherpa-onnx contributors
Note: our shipped silero_vad.onnx is byte-identical to the file served by
  csukuangfj/vad (sha256 a35ebf52fd3ce5f1469b2a36158dba761bc47b973ea3382b3186ca15b1f5af28).
  The repackaging does not relicense the weights, so the governing license is the
  upstream MIT. This file previously claimed "CC-BY-NC-4.0 (original)"; we could not
  find any such grant for Silero VAD - the repository's own license metadata says MIT -
  so the claim and its appendix entry were removed rather than copied forward.

--- SenseVoice (Speech Recognition + Emotion Detection) ---
License: FunASR Model Open Source License Agreement v1.1 (not Apache-2.0)
Source: csukuangfj/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17 (Hugging Face;
  int8 ONNX conversion; the card declares no license tag)
Original: Alibaba FunAudioLLM / iic SenseVoiceSmall
Copyright (C) 2023-2028 Alibaba Group. All rights reserved.
URL: https://github.com/modelscope/FunASR/blob/main/MODEL_LICENSE
License conflict, recorded rather than resolved (both read 2026-09-25):
  - the Hugging Face card (FunAudioLLM/SenseVoiceSmall) sets license: other,
    license_name: model-license, license_link -> FunASR/MODEL_LICENSE;
  - the same model's ModelScope metadata (iic/SenseVoiceSmall) reports
    "Apache License 2.0" with an empty LicenseLink, and the repository ships no
    LICENSE file.
  We comply with the stricter instrument (the FunASR agreement, whose only real
  duty is 2.2 attribution + name retention) because that is what the authoring
  org wrote on the card, and Apache-2.0 grants strictly less obligation, not more
  permission than the card claims. If counsel ever needs one answer, this is the
  item to ask about.
Obligations we must keep (verified against the agreement text, 2026-09-25):
  - 2.1 use / copy / modify / share are granted, including commercially;
  - 2.2 you MUST attribute the source and author information and RETAIN THE MODEL
    NAMES. That is why this entry, the model's on-disk directory name and the
    in-app model list all keep the "SenseVoice" name rather than a generic label;
  - there is no warranty and Alibaba Group assumes no liability (section 3).
Note: an earlier revision of this file said "Apache-2.0". That was wrong - the
  upstream card is deliberately tagged "other" because it is this custom agreement.
  See the LICENSE TEXTS appendix for the agreement location.

--- Whisper GGML Models (base / small / large-v3-turbo) ---
License: MIT
Source: whisper.cpp (https://github.com/ggerganov/whisper.cpp)
Original: OpenAI Whisper (MIT)
Copyright (c) Georgi Gerganov (whisper.cpp) / OpenAI (Whisper)

--- Whisper MLX Models (base / small / large-v3-turbo) ---
License: MIT
Source: mlx-community/whisper-*-mlx (Hugging Face)
Original: OpenAI Whisper (MIT)
Copyright (c) MLX Community / OpenAI (Whisper)

--- Whisper MLX Assets (multilingual tokenizer, GPT2 tokenizer, mel filters) ---
License: MIT
Source: Bundled with whisper-MLX, derived from OpenAI Whisper
Copyright (c) MLX Community / OpenAI

--- Qwen3-TTS-ONNX (0.6B, fp16) ---
License: Apache-2.0
Source: onnx-community/Qwen3-TTS-12Hz-0.6B-Base (Hugging Face; card tag license:apache-2.0)
Original: Qwen/Qwen3-TTS-12Hz-0.6B-Base (cardData license: apache-2.0, read 2026-09-25)
Copyright (c) Alibaba Qwen Team
URL: https://huggingface.co/Qwen/Qwen3-TTS-12Hz-0.6B-Base
Note: the upstream repository ships no LICENSE file, so the grant rests on the card
  metadata. This entry previously parenthesised it as "(Qwen Research License)"; the
  0.6B base card says plain Apache-2.0, and the larger Qwen models are the ones under
  the Qwen Research / Qwen license agreements - we do not ship those.

--- Qwen3-TTS-MLX (0.6B, 8-bit, Apple Silicon) ---
License: Apache-2.0
Source: mlx-community/Qwen3-TTS-12Hz-0.6B-Base-8bit (Hugging Face; card tag license:apache-2.0)
Original: Alibaba Qwen Team (Apache-2.0)
Copyright (c) MLX Community / Alibaba Qwen Team

--- MOSS-TTS-Nano-100M-ONNX-fp16 (GPU accelerated) ---
License: Apache-2.0
Source: REALBITS/MOSS-TTS-Nano-100M-ONNX-fp16 (Hugging Face)
Original: OpenMOSS Team (Apache-2.0)
Copyright (c) REALBITS / OpenMOSS Team

--- Subtitle Removal - LaMa Inpainting (ONNX INT8, Windows/CPU) ---
License: Apache-2.0 at every hop of the chain
Source: opencv/inpainting_lama (Hugging Face) - the repository ships its own LICENSE
  file with the full Apache-2.0 text and its README states "All files in this
  directory are licensed under Apache License"
Original: advimman/lama (GitHub, Apache-2.0 per the repository license metadata);
  the opencv card names Carve/LaMa-ONNX (card tag license:apache-2.0) as the model source
Copyright (c) LaMa authors (advimman et al.) / OpenCV contributors
Verification (2026-09-25): our shipped inpainting_lama_2025jan.onnx is byte-identical
  to the file of the same name in opencv/inpainting_lama
  (sha256 7df918ac3921d3daf0aae1d219776cf0dc4e4935f035af81841b40adcf74fdf2).
  An earlier revision of this file said "S3 distribution (no public HF repo)"; that was
  wrong - S3/R2 is only our fallback tier, the public origin is the repo above.

--- Subtitle Removal - LaMa Inpainting (MLX BF16, Apple Silicon) ---
License: Apache-2.0
Source: mlx-community/LaMa-bf16 (Hugging Face; card tag license:apache-2.0)
Original: advimman/lama (Apache-2.0)
Copyright (c) MLX Community / LaMa authors

--- Subtitle Removal - DBNet v3 Text Detection (det_v3.onnx) ---
License: MIT (converter card) over Apache-2.0 (the underlying PaddleOCR model)
Source: alterf/det_v3 (Hugging Face; card tag license:mit, read 2026-09-25)
Original: PaddleOCR / PP-OCRv3 mobile text detection (Baidu, Apache-2.0)
Copyright (c) Baidu PaddlePaddle contributors / alterf
Note: a 2.4 MB text-detection graph; the ONNX export adds no new rights claim
  beyond the upstream Apache-2.0 model it converts.

--- PP-OCRv5 (Watermark Text Detection + Recognition + Dictionary) ---
License: Apache-2.0 at every hop
Source: ilaylow/PP_OCRv5_mobile_onnx (Hugging Face; card tag license:apache-2.0)
        for det.onnx + rec.onnx; the dictionary ships from
        bluecopa/paddleocr-v5-onnx (its card also declares apache-2.0)
Original: PaddleOCR / PP-OCRv5 mobile (Baidu, Apache-2.0)
Copyright (c) Baidu / PaddlePaddle contributors
Verification (2026-09-25): our det_mobile.onnx and rec_mobile.onnx are byte-identical
  to ilaylow's, which are in turn byte-identical to bluecopa's
  (sha256 1eb7b4f7ab657ebd... / 243a0f06d8267613...); the dictionary is byte-identical
  to bluecopa's (sha256 d1979e9f794c464c..., 74,012 B).
  An earlier revision said "no public HF repo" - wrong; and it listed bluecopa as the
  only source, which cannot serve det/rec under the names we check (bluecopa ships the
  *server* graphs). Because no single repo serves all three check_files, the dictionary
  download always falls through to the R2 tier - see the registry entry's comment.

--- Mel-Band RoFormer Vocal Separation (ONNX int8 4s window + MLX fp32) ---
License: MIT at every hop
Source: we distribute our own derivatives (int8 dynamic quantization, and a 4s-window
  graph edit), built from smank/mel-band-roformer-vocals-onnx fp32
  (Hugging Face; card tag license:mit)
Original chain declared by that repository's README: the Vocals checkpoint from
  KimberleyJSN/melbandroformer (card tag license:mit), the RoFormer/Mel-Band RoFormer
  architecture from ZFTurbo (MIT, arXiv:2310.01809), and the mixxxdj/demucs baseline
  (MIT)
Copyright (c) KimberleyJSN / ZFTurbo / mixxxdj contributors / smank
Outstanding risk (disclosed, not resolved by license text): none of the cards above
  publish a training-corpus manifest. The checkpoints are released as MIT, which grants
  use and commercial use, but the provenance of the audio they were trained on is not
  documented upstream. See section 8.

--- Chatterbox Multilingual TTS (ONNX + MLX LM companion) ---
License: MIT at every hop we can see
Source: onnx-community/chatterbox-multilingual-ONNX (Hugging Face; card tag
  license:mit, base_model declared as ResembleAI/chatterbox)
Original: ResembleAI/chatterbox (card tag license:mit)
Copyright (c) Resemble AI / onnx-community contributors
MLX companion: language_model_mlx_w8.safetensors is our own 8-bit quantization of the
  fp16 LM inside the same ONNX repository and inherits its MIT grant; it is served only
  from our R2 tier because upstream has no such file.
Outstanding risk (disclosed, not resolved by license text): the MIT grant covers the
  weights as published; Resemble AI does not publish the training corpus or the
  provenance/licence of any speaker audio in it. For a product that clones voices this
  is the single largest legal unknown in this file - see section 8.

================================================================================
4. RUST CRATES (Selected)
================================================================================

All Rust dependencies use permissive licenses (MIT, Apache-2.0, ISC, BSD, Unicode).
Key dependencies:

| Crate                        | License           | Copyright                    |
|------------------------------|-------------------|------------------------------|
| tauri + plugins              | MIT / Apache-2.0  | Tauri Contributors           |
| serde / serde_json           | MIT / Apache-2.0  | David Tolnay                 |
| tokio / tokio-util           | MIT               | Tokio Contributors           |
| reqwest                      | MIT / Apache-2.0  | Sean McArthur                |
| aws-sdk-s3, aws-config       | Apache-2.0        | Amazon Web Services          |
| aws-lc-rs, aws-lc-sys        | ISC               | AWS Cryptography             |
| axum                         | MIT               | Tokio Contributors           |
| tower-http                   | MIT               | Tower Contributors           |
| sentry                       | MIT               | Sentry Contributors          |
| chrono                       | MIT / Apache-2.0  | Chrono Contributors          |
| uuid                         | MIT / Apache-2.0  | Rust UUID Contributors       |
| log                          | MIT / Apache-2.0  | Rust Developers              |
| tracing-subscriber           | MIT               | Tokio Contributors           |
| tracing-appender             | MIT               | Tokio Contributors           |
| tracing-log                  | MIT / Apache-2.0  | Tokio Contributors           |
| font-kit                     | MIT / Apache-2.0  | Patrick Walton               |
| image                        | MIT               | Image Contributors           |
| hound                        | MIT / Apache-2.0  | Tadej Janež                  |
| keyring                      | MIT / Apache-2.0  | Keyring Contributors         |
| open                         | MIT / Apache-2.0  | open Contributors            |
| dirs                         | MIT / Apache-2.0  | dirs Contributors            |
| base64                       | MIT / Apache-2.0  | Alice Maz                    |
| hmac, sha2                   | MIT / Apache-2.0  | RustCrypto                   |
| objc, cocoa, objc_exception  | MIT               | objc Contributors            |
| security-framework           | MIT / Apache-2.0  | Steven Fackler               |
| windows-sys                  | MIT / Apache-2.0  | Microsoft                    |
| wetext-rs                    | Apache-2.0        | SpenserCai / WeNet Contributors |
| rustfst                      | MIT / Apache-2.0  | RustFST Contributors         |
| ort                          | MIT               |ORT Contributors               |
| whisper-rs / whisper-rs-sys  | MIT               | Tegan Milligan               |
| sherpa-onnx / sherpa-onnx-sys| Apache-2.0        | sherpa-onnx contributors     |
| mlx-rs                       | MIT               | MLX-RS Contributors          |
| ndarray                      | MIT / Apache-2.0  | ndarray Contributors         |
| rubato                       | BSD-3-Clause      | Rubato Contributors          |
| rustfft                      | MIT / Apache-2.0  | Allen Welton                 |
| sentencepiece-rs             | Apache-2.0        | sentencepiece-rs Contributors|
| half                         | MIT / Apache-2.0  | half Contributors            |
| num-complex                  | MIT / Apache-2.0  | Rust Numerics Contributors   |
| npyz                         | MIT / Apache-2.0  | npyz Contributors            |
| futures-util / async-stream  | MIT / Apache-2.0  | Rust Async Contributors     |
| once_cell                    | MIT / Apache-2.0  | Aleksey Kladov               |
| rand                         | MIT / Apache-2.0  | Rust RNG Contributors        |
| libloading                   | ISC               | Simonas Kazlauskas           |

A complete list of all transitive Rust crate dependencies with their
license identifiers is recorded in Cargo.lock in the source repository.
Full license texts for any specific crate can be obtained from:
  https://crates.io/crates/<crate-name>/<version>

================================================================================
5. JAVASCRIPT / FRONTEND
================================================================================

--- @tauri-apps/api ---
License: MIT
Copyright (c) Tauri Contributors
URL: https://github.com/tauri-apps/tauri

The frontend uses hand-written vanilla JavaScript with no additional
third-party npm dependencies.

================================================================================
6. FFMPEG (External Binary)
================================================================================

FFmpeg is an external binary used for audio/video processing. It runs as a
separate process spawned by the application; no FFmpeg code is linked into the
application binary.

One binary is bundled per platform, and they are NOT the same build. The
configuration lines below were read out of the shipped files themselves on
2026-09-25 (`ffmpeg -version`, or the embedded configuration string), not from
a build note.

  Copyright (c) 2000-2025 the FFmpeg developers
  URL: https://ffmpeg.org

--------------------------------------------------------------------------------
6.1 macOS arm64 (bundled, Apple Silicon) — LGPL v2.1 or later
--------------------------------------------------------------------------------

Built by us from unmodified upstream FFmpeg 7.1.1 source.

  configuration: --disable-gpl --disable-nonfree --disable-version3
    --enable-static --disable-shared --enable-videotoolbox
    --enable-audiotoolbox --enable-libass --enable-libfreetype
    --enable-libfontconfig --enable-libfribidi --enable-libharfbuzz
    --disable-xlib --disable-sdl2 --disable-avdevice --disable-postproc
    --extra-cflags='-mmacosx-version-min=14.0 -O3'

No GPL component is present, which is why the H.264 encoder used for watermark
and subtitle burn-in is `h264_videotoolbox` (Apple's hardware encoder) rather
than `libx264`.

FFmpeg is free software; you can redistribute it and/or modify it under the
terms of the GNU Lesser General Public License as published by the Free
Software Foundation; either version 2.1 of the License, or (at your option)
any later version.

FFmpeg is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
details.

The corresponding source for this build (upstream 7.1.1 plus the build script
in src-tauri/third_party/ffmpeg-lgpl/) is available on request at
deepvideo@fooodiee.com, and is otherwise published at
https://ffmpeg.org/download.html

Because the FFmpeg libraries are statically linked into the ffmpeg executable,
the LGPL's relinking requirement is met by the written offer above. The
external libraries it loads at runtime are shipped as replaceable .dylib files
next to the binary, so a user can substitute their own build.

  Bundled dynamic dependencies (macOS arm64), all in Contents/MacOS/:

    libass.9.dylib        ISC
    libfontconfig.1.dylib HPND-sell-variant, Unicode-3.0, MIT-Modern-Variant,
                          MIT, public-domain (data tables)
    libfreetype.6.dylib   FTL (FreeType License, BSD-style)
    libfribidi.0.dylib    LGPL-2.1-or-later
    libglib-2.0.0.dylib   LGPL-2.1-or-later
    libgraphite2.3.dylib  MIT
    libharfbuzz.0.dylib   MIT
    libintl.8.dylib       LGPL-2.1-or-later (gettext runtime library)
    libpcre2-8.0.dylib    BSD-3-Clause
    libpng16.16.dylib     libpng-2.0
    libunibreak.7.dylib   Zlib

  Upstream projects are listed at https://ffmpeg.org/download.html and in each
  library's own repository; their license texts accompany the source offer
  above.

--------------------------------------------------------------------------------
6.2 macOS x86_64 (bundled, Intel) — third-party build, NOT LGPL
--------------------------------------------------------------------------------

  configuration (as embedded in the shipped file):
    --prefix=/Volumes/tempdisk/sw --arch=x86_64 --enable-videotoolbox
    --enable-libfreetype --enable-fontconfig --enable-gpl --enable-libvmaf
    --enable-libopenjpeg --enable-libopus --enable-libtheora
    --enable-libvorbis --enable-libmp3lame --enable-libx264 --enable-libx265
    --enable-libvpx --enable-libwebp --enable-libass --enable-libaom
    --enable-libsnappy --enable-libvidstab --enable-libzimg
    --enable-libsvtav1 --enable-libharfbuzz --enable-version3
    --enable-postproc --enable-nonfree

This build carries --enable-gpl, --enable-version3 and --enable-nonfree, so it
is subject to the GNU General Public License v3 (or later) as extended by the
version-3 components, and the --enable-nonfree parts additionally restrict
redistribution. It is therefore also unusable for Mac App Store submission.

Disclosed here because it is a known defect, not a settled position: replacing
this binary with our own LGPL build (same script, `--arch x86_64`) is tracked
as an open item. Until then, Intel users receive the binary described in this
subsection, and its source/offer obligations are those of GPL v3 as published
by the Free Software Foundation; a request for the corresponding source can be
sent to deepvideo@fooodiee.com.

--------------------------------------------------------------------------------
6.3 Windows x64 (bundled) — GPL v3
--------------------------------------------------------------------------------

Third-party build from gyan.dev (FFmpeg "full" build), verified from the
configuration embedded in the bundled ffmpeg.exe:

  configuration: --enable-gpl --enable-version3 --enable-static
    --enable-cairo --enable-fontconfig --enable-gnutls --enable-libxml2
    --enable-libsrt --enable-libssh --enable-libzmq --enable-avisynth
    --enable-libwebp --enable-libx264 --enable-libx265 --enable-libxvid
    --enable-libaom --enable-libopenjpeg --enable-libvpx --enable-mediafoundation
    --enable-libass --enable-libfreetype --enable-libfribidi
    --enable-libharfbuzz --enable-libvidstab --enable-libvmaf --enable-libzimg
    --enable-amf --enable-cuvid --enable-dxva2 --enable-d3d11va
    --enable-nvenc --enable-libmp3lame and others.

FFmpeg is free software; you can redistribute it and/or modify it under the
terms of the GNU General Public License as published by the Free Software
Foundation; either version 3 of the License, or (at your option) any later
version. A copy of the GPL v3 is available at
https://www.gnu.org/licenses/gpl-3.0.txt, and the corresponding source for
this build is published at https://www.gyan.dev/ffmpeg/builds/ and
https://ffmpeg.org/download.html. A written offer can be sent to
deepvideo@fooodiee.com.

Windows is not distributed through the Mac App Store, so the GPL terms are met
by the source offer above rather than by removing the GPL encoders.

--------------------------------------------------------------------------------
6.4 Linux
--------------------------------------------------------------------------------

On Linux, FFmpeg is installed as a system package (not bundled); the .deb
declares `ffmpeg` as a dependency, so the distro's own licensing applies.

--------------------------------------------------------------------------------
6.5 Historical note
--------------------------------------------------------------------------------

Earlier releases obtained FFmpeg through the imageio-ffmpeg Python wrapper
(BSD-2-Clause wrapper around a gyan.dev build) on Windows and macOS. That
wrapper is no longer part of the product: Python was removed, and the binaries
described above are bundled directly.

================================================================================
7. OTHER TOOLS
================================================================================

--- yt-dlp (Video Download, optional external tool) ---
License: Unlicense (Public Domain)
URL: https://github.com/yt-dlp/yt-dlp

--- Cloudflare R2 (Object Storage CDN) ---
Accessed via S3-compatible API with read-only credentials.
Service governed by Cloudflare Terms of Service.

================================================================================
8. MODEL TRAINING-DATA DISCLOSURES (what the licenses above do NOT settle)
================================================================================

Every license in section 3 settles the right to REDISTRIBUTE A PUBLISHED
WEIGHT FILE. None of them is a statement about what that file was trained on.
For a product whose core feature is cloning a voice from 15 seconds of
reference audio, that distinction is the residual risk, so it is recorded here
rather than hidden behind a permissive license badge.

Status column meaning: VERIFIED = we read the upstream text/repo ourselves on
2026-09-25; AGENT-SOURCED = a research subagent reported it and we did not
re-read it; UNKNOWN = upstream simply does not document it.

--- 8.1 Chatterbox Multilingual (Resemble AI) - UNKNOWN, highest exposure ---
The MIT card grants use of the weights. Resemble AI publishes no training
corpus manifest and no licence or consent record for the speaker audio behind
the model, and the repository also ships a built-in `default_voice.wav` whose
recording provenance we cannot see. Consequences to weigh before marketing
work: voice-cloning output can resemble a real person, and a permissive weight
licence does not immunise a customer (or us) from personality-rights,
biometric or deepfake rules in the target jurisdiction. Mitigation we do
control: the reference audio is user-supplied, and the EULA puts the consent
burden on them; the disclosure duty is on us.

--- 8.2 Mel-Band RoFormer vocals (smank / KimberleyJSN / ZFTurbo) - UNKNOWN ---
Three hops, all MIT, none of them declaring a dataset. A vocal-separation
checkpoint is by construction fitted to recorded music, which points at
licensed corpora (the community default is MUSDB18-HQ, whose terms are for
research/benchmark use and are not a commercial licence for the underlying
tracks). We verified the licence tags, not the training data. The weights are
used only as an internal step (diarisation input), never redistributed as music.

--- 8.3 htdemucs / htdemucs-mlx (Meta, MIT) - DOCUMENTED RESIDUE, read 2026-09-25 ---
Unlike 8.2 this one IS documented upstream: the Demucs README states plainly
"It has been trained on the MUSDB HQ dataset + an extra training dataset of
800 songs", and its license section says "Demucs is released under the MIT
license as found in the LICENSE file". So Meta grants the weights under MIT,
but it also tells us MUSDB18-HQ is in the mix - and MUSDB18-HQ's own terms
cover research/benchmark use of stems whose underlying recordings stay with
their labels. Meta's MIT grant cannot convey those recordings to us. Same
posture as 8.2: an internal processing step, never redistribution of music.

--- 8.4 SenseVoice (Alibaba) - VERIFIED, duty is active ---
FunASR MODEL_LICENSE v1.1 section 2.2 requires that we attribute the source and
author information AND retain the model names. This file, the model directory
name and the in-app model list keep the "SenseVoice" name for that reason.
Removing or genericising that label would breach the grant, not merely look
worse.

--- 8.5 3D-Speaker ERes2Net (Alibaba DAMO) - PARTIAL ---
Apache-2.0 on the card we verified. The 3D-Speaker project's own training
corpora (VoxCeleb, CN-Celeb, and the team's internal data) carry dataset-level
terms: VoxCeleb's use conditions include a non-commercial clause for some
derivatives and a citation duty. Speaker-embedding models are the closest thing
in this product to biometric processing, so both the dataset terms and
biometric consent (again the user's burden for the audio they upload) are worth
naming to counsel.

--- 8.6 pyannote segmentation-3.0 (MIT, via sherpa-onnx) - VERIFIED TAG ---
The csukuangfj converter card declares Apache-2.0; the original
pyannote/segmentation-3.0 card declares MIT. Neither publishes a training
corpus. The heavier pyannote community-1 pipeline, which is CC-BY-4.0 and was
gated on Hugging Face, is NOT shipped and its registry entry was deleted; an
earlier revision of this file listed it, which was misleading.

--- 8.7 Whisper (OpenAI MIT) - KNOWN, not settled by the licence ---
OpenAI's MIT grant covers the weights. The model card states the training data
included audio scraped from public web sources under "representative of English"
sampling; OpenAI does not publish the manifest. Distinct from the licence
question, and identical in kind to every other public ASR model we examined.

--- 8.8 What we did NOT verify ---
- Windows CPU peak memory for any candidate model discussed in the 2026-09-25
  selection review: no Windows machine was available, so those numbers are
  AGENT-SOURCED estimates from vendor documentation, not measurements.
- The smank fp32 source ONNX hash: our shipped 4s-window int8 file is derived
  from it, but we did not re-download the 953 MB fp32 original to prove the
  lineage byte-for-byte; the derivation script and the write-back comparison
  (weights identical after restoring the 8s window constant) are the evidence.
- Whether the ilaylow / bluecopa PP-OCRv5 conversion cards' Apache-2.0 tag
  binds Baidu's own model, or only the converter's contribution. Legally the
  upstream PaddleOCR Apache-2.0 is what matters and is the reason we can ship
  it either way.

================================================================================
LICENSE TEXTS
================================================================================

--- 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.

--- Apache License 2.0 ---

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

--- BSD 3-Clause License ---

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice,
   this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice,
   this list of conditions and the following disclaimer in the documentation
   and/or other materials provided with the distribution.
3. Neither the name of the copyright holder nor the names of its contributors
   may be used to endorse or promote products derived from this software
   without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
POSSIBILITY OF SUCH DAMAGE.

--- BSD 2-Clause License ---

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice,
   this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice,
   this list of conditions and the following disclaimer in the documentation
   and/or other materials provided with the distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
POSSIBILITY OF SUCH DAMAGE.

--- GNU General Public License v3.0 (GPL-3.0) ---

Full text available at: https://www.gnu.org/licenses/gpl-3.0.txt

--- GNU Affero General Public License v3.0 (AGPL-3.0) ---

Full text available at: https://www.gnu.org/licenses/agpl-3.0.txt

--- FunASR Model Open Source License Agreement v1.1 ---

Governs the SenseVoice model we ship for ASR + emotion detection.
Full text: https://github.com/modelscope/FunASR/blob/main/MODEL_LICENSE
(Copyright (C) 2023-2028 Alibaba Group; use/copy/modify/share including
commercial use are granted, subject to the attribution duty in 2.2.)

--- Mozilla Public License 2.0 (MPL-2.0) ---

Full text available at: https://www.mozilla.org/en-US/MPL/2.0/

================================================================================
END OF THIRD-PARTY LICENSES
================================================================================
