hex/THIRD_PARTY_NOTICES.md
Dax 4d28a692d8
add fluid speech pro to the speech model catalog (#2007)
* add fluid speech pro to the speech model catalog

FluidVoice's Fluid Speech Pro, a Parakeet TDT 0.6B v3 retrain that also
declares Moondream's Parakeet Ultra as a base, for the same 25 European
languages. TalkMate converted it to a Q8_0 transcribe.cpp GGUF at d59dc79
(741,819,456 bytes, sha256 c3848e98...); the file says
parakeet/tdt-0.6b-fluid-speech-pro and announces its extra pre-Conformer
block as stt.parakeet.encoder.pre_block_kernel, so it needs a transcribe.cpp
that reads enc.pre_block.

Like Parakeet v3 it takes no language hint, detects the language under Auto,
does not stream, and is never recommended or badged. Its card publishes no
LibriSpeech test-clean figure, so it carries no single word error rate and no
benchmark record; its per-language FLEURS figures and the CC BY 4.0 credit
are in its details.

* bump(transcribe.cpp): 295c7e6, fluid speech pro's pre-conformer block

Fluid Speech Pro's GGUF announces stt.parakeet.encoder.pre_block_kernel and
carries eight enc.pre_block.* tensors: a residual FrameCleaner block between
the subsampler and Conformer layer 0. The old pin's graph never read them, so
the model loaded as plain Parakeet v3 and its encoder was wrong.

Both transcribe-cpp crates move to 295c7e601 on hex/combined-fluid, a
fast-forward of bcb639f that runs the block on pre_encode before xscaling
when the KV is present, and fails the load rather than skipping tensors when
it is only partly there. Cargo.lock changes only those two sources
(`getrandom`'s tempfile pointer was reverted to what it was), and
nix/package.nix carries the hash from a failed `nix build .#hex`, then a full
one whose check phase passed.

`hex model install fluid-speech-pro` followed by `hex model transcribe` on
whisper.cpp's jfk.wav prints the expected sentence, as does Parakeet v3 on
the same clip.

* fluid speech pro: snake_case id, FLEURS-only figures, blurb, notices
2026-11-19 15:19:06 -05:00

12 KiB
Raw Permalink Blame History

Third-Party Notices

Hex's own source is covered by the root LICENSE. The following notices cover vendored icon assets and the speech models Hex downloads under their own licences. Dependencies retain their respective upstream licenses.

Parakeet Ultra

Parakeet Ultra by Moondream, based on NVIDIA Parakeet TDT 0.6B v3. GGUF conversion and quantization by handy-computer. Creative Commons Attribution 4.0 International (CC BY 4.0). Hex downloads parakeet-ultra-Q8_0.gguf from huggingface.co/handy-computer/parakeet-ultra-gguf@a1539be1b8 only while Labs › Parakeet Ultra is on; it does not ship the weights.

License: https://creativecommons.org/licenses/by/4.0/legalcode.en

Notice supplied with the files:

Modifications: handy-computer converted https://huggingface.co/moondream/parakeet-ultra at commit 73175eb to GGUF for transcribe.cpp (https://github.com/handy-computer/transcribe.cpp). Conversion repackages the weights, renaming and re-laying-out tensors; the quantized files (Q8_0, Q6_K, Q5_K_M, Q4_K_M) also store them at reduced numeric precision. The weights were not retrained or fine-tuned.

The weights are provided as-is, without warranties, as the licence states.

Orukeet

Orukeet by Oruk AI, based on NVIDIA Parakeet TDT 0.6B v3. GGUF conversion and quantization by handy-computer. Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0). Hex downloads orukeet-Q8_0.gguf from huggingface.co/handy-computer/orukeet-gguf@c5511a3850 and does not ship the weights.

License: https://creativecommons.org/licenses/by-sa/4.0/legalcode.en

Notice supplied with the files:

Attribution and license scope

Orukeet is an adaptation of NVIDIA Parakeet TDT 0.6B v3. NVIDIA retains copyright in its model and upstream work. The base weights are distributed under CC BY 4.0. Oruk AI's changes comprise multilingual/accent continuation training, parameter blending, fitted and frozen Gabor-kernel replacement and recovery, native and ONNX export, application integration and evaluation. Model stages identifies r3 as the source of every current Orukeet download and records its ancestry.

The r3 NeMo, ONNX INT8, Q8 GGUF and F16 GGUF weights, and their fitted Gabor kernels, are designated CC BY-SA 4.0. The complete license is in LICENSE-WEIGHTS. This permits commercial use and modification, with attribution and applicable ShareAlike requirements. Earlier checkpoints retain their source notices and are not silently relicensed by this file. Orukeet v0.1.0 distributes the final r3 checkpoint in all four formats.

Python and integration code in this repository is MIT unless a file specifies otherwise. The native bindings, audio windowing and worker transport derive from OpenWhisper by Knuckles92 and its Oruk AI integration; their MIT notice is retained in LICENSE. The downloaded NVIDIA NeMo-Speech.cpp SDK and its bundled dependencies retain their own notices, including Apache-2.0 and MIT components. Keep the SDK's license files when redistributing it. The pinned source is the authority for those terms.

The ONNX exporter follows sherpa-onnx's Parakeet TDT v3 conversion script. Its upstream reference and Apache-2.0 license are retained in export/onnx. The ONNX archive includes the weight license and source attribution. The optimized ONNX encoder uses exactly equivalent FP32 arithmetic for 24 quantized depthwise convolutions, with existing runtime operators and unchanged quantized values. Original export and separate optimization receipts are retained in the GitHub repository under evidence/onnx-r3-20260910/ and evidence/speed20260910/.

Training data credits:

  • Mozilla/Common Voice contributors: Common Voice 22, CC0.
  • Google and the FLEURS authors: FLEURS, CC BY 4.0.
  • Google and the OpenSLR 83 authors, with the ylacombe/english_dialects restructuring: English dialect speech, CC BY-SA 4.0.
  • Beijing Kingline Data Technology and the SpeechOcean762 authors: SpeechOcean762, CC BY 4.0.
  • SpeechColab and the GigaSpeechBench authors: 17 English accent/domain splits in the final continuation. Their paper identifies Creative Commons source audio; the source record below preserves the available license details.
  • DISCO at ETH Zurich and the EuroSpeech contributors: Bulgarian, Greek and Italian parliamentary speech, with the providers' country-specific terms.
  • SberDevices and the Golos authors: Golos Crowd, under the Public license with attribution and conditions reserved.
  • Nordisk Språkteknologi, the National Library of Norway and the Alexandra Institute: NST Swedish and Danish, distributed under CC0.
  • ILSP/Athena Research Center and the Lesbian Speech Corpus contributors: dialect speech from Lesbos. The source card does not specify a reuse license.

Data were filtered, split, normalized and sampled as described in data and licenses. Corpus copyrights and source terms remain with their owners. Evaluation-only data have separate terms. No endorsement by NVIDIA, Mozilla, Google or the other source projects is implied.

The generated statistical records in evidence/metric-evidence.tar.gz, the Gabor recovery count bundles under training/gabor_half/results/, and the FT-4035 benchmark counts under evidence/regression-ft-20260907/ are designated CC BY 4.0, attributed to Oruk AI. This designation covers the prepared metric records; it does not relicense the original corpus recordings or transcripts, which are not included in that archive.

The r3 numeric benchmark records in evidence/standard-asr-r3-20260908/ and evidence/domains-r3-20260908/ are released under CC BY 4.0. They contain error counts and recording identifiers; dataset audio remains with its providers.

The paired ONNX application benchmark counts in evidence/onnx-r3-20260910/ are also released under CC BY 4.0. These records contain numeric errors, timings and recording identifiers, without corpus audio or transcripts.

Modifications: handy-computer converted https://huggingface.co/oruk/orukeet at commit b59c13a to GGUF for transcribe.cpp (https://github.com/handy-computer/transcribe.cpp). Conversion repackages the weights, renaming and re-laying-out tensors; the quantized files (Q8_0, Q6_K, Q5_K_M, Q4_K_M) also store them at reduced numeric precision. The weights were not retrained or fine-tuned.

The weights are provided as-is, without warranties, as the licence states.

Fluid Speech Pro

Fluid Speech Pro by FluidVoice, based on NVIDIA Parakeet TDT 0.6B v3 and Moondream Parakeet Ultra. GGUF conversion and quantization by TalkMate. Creative Commons Attribution 4.0 International (CC BY 4.0). Hex downloads fluid-speech-pro-Q8_0.gguf from huggingface.co/TalkMate/fluid-speech-pro-gguf@d59dc79f7e and does not ship the weights.

Copyright (c) 2026 FluidVoice. Licensed under CC BY 4.0 (see LICENSE).

Attribution for use: "Fluid Speech Pro by FluidVoice. Built on NVIDIA Parakeet TDT 0.6B v3 and Moondream Parakeet Ultra."

License: https://creativecommons.org/licenses/by/4.0/legalcode.en

Notice supplied with the files:

Fluid Speech Pro by FluidVoice (https://huggingface.co/FluidVoice-Labs/fluid-speech-pro), CC BY 4.0. Built on NVIDIA Parakeet TDT 0.6B v3 and Moondream Parakeet Ultra, both CC BY 4.0. Converted to a transcribe.cpp GGUF (Q8_0) by TalkMate, in the layout of Handy's Parakeet Ultra GGUF (https://huggingface.co/handy-computer/parakeet-ultra-gguf, CC BY 4.0).

The weights are provided as-is, without warranties, as the licence states.

Wayland Input Method Protocol

src/linux/protocols/input-method-unstable-v2.xml is the upstream input-method-unstable-v2 protocol description, copied unchanged from wayland-protocols-misc 0.3.12. It is MIT licensed; the copyright header is kept in the file.

Upstream: https://gitlab.freedesktop.org/wayland/wayland-protocols

Lucide And Feather

The settings sidebar icons in assets/icons/ (settings, modes, meetings, history, voice-filter, scrawl, hud-lab, activity) are Lucide 1.62.0's sliders-horizontal, layers, users, history, audio-waveform, scan-line, flask-conical, and activity, with stroke attributes moved onto each element and a 1.75 stroke. The assets/icons/lucide/*.svg assets in the Windows development branch are also from Lucide: https://github.com/lucide-icons/lucide

Upstream license: github.com/lucide-icons/lucide@23f9abc4ed/LICENSE

search.svg has a modified handle endpoint and reordered SVG elements. The complete upstream license below covers both original and derived assets.

ISC License

Copyright (c) 2026 Lucide Icons and Contributors

Permission to use, copy, modify, and/or distribute this software for any purpose with or without fee is hereby granted, provided that the above copyright notice and this permission notice appear in all copies.

THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.


The following Lucide icons are derived from the Feather project:

airplay, alert-circle, alert-octagon, alert-triangle, aperture, arrow-down-circle, arrow-down-left, arrow-down-right, arrow-down, arrow-left-circle, arrow-left, arrow-right-circle, arrow-right, arrow-up-circle, arrow-up-left, arrow-up-right, arrow-up, at-sign, calendar, cast, check, chevron-down, chevron-left, chevron-right, chevron-up, chevrons-down, chevrons-left, chevrons-right, chevrons-up, circle, clipboard, clock, code, columns, command, compass, corner-down-left, corner-down-right, corner-left-down, corner-left-up, corner-right-down, corner-right-up, corner-up-left, corner-up-right, crosshair, database, divide-circle, divide-square, dollar-sign, download, external-link, feather, frown, hash, headphones, help-circle, info, italic, key, layout, life-buoy, link-2, link, loader, lock, log-in, log-out, maximize, meh, minimize, minimize-2, minus-circle, minus-square, minus, monitor, moon, more-horizontal, more-vertical, move, music, navigation-2, navigation, octagon, pause-circle, percent, plus-circle, plus-square, plus, power, radio, rss, search, server, share, shopping-bag, sidebar, smartphone, smile, square, table-2, tablet, target, terminal, trash-2, trash, triangle, tv, type, upload, x-circle, x-octagon, x-square, x, zoom-in, zoom-out

The MIT License (MIT) (for the icons listed above)

Copyright (c) 2013-present Cole Bemis

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.