Turns English speech in your video into timed subtitles on your own computer, with Phonon-2, the speech model from Fermion Research. It runs on the CPU, needs no graphics card and no account, and nothing leaves your machine.

What it does

Offline after one download

The 164 MB model downloads once, the first time you transcribe, and is checked against a pinned fingerprint. From then on, no internet is needed.

No graphics card

It runs on the CPU. Set how many cores it may use, or leave it to choose.

Subtitles that read well

Phonon-2 cuts speech at sentences and pauses, so each subtitle starts and ends where the speaker does.

Your names, spelled right

List the names and terms your video uses under Hotwords, and Phonon-2 favours them.

Models

There is one model, Phonon-2. It is not inside the add-on: it downloads into Subtitld's shared model cache the first time you transcribe, so updating the add-on doesn't download it again.

Model	Download	Good for
Phonon-2 Default	164 MB	English speech, from interviews to lectures

The first transcription also loads the model, so it takes longer than the ones after it. Phonon-2 transcribes a file in one pass: Cancel takes effect once that pass ends.

Options

In Settings › Add-ons › Phonon-2 (offline, English) › Configure.

Models folder

Default: empty

Where the model is kept. Leave it empty to share Subtitld's model cache, so updating or removing the add-on keeps the model.

CPU threads

Default: 0

How many CPU cores Phonon-2 uses. 0 picks for you. A new value applies the next time the model loads.

Hotwords

Default: empty

Names and terms from your video, separated by commas, up to 25. Phonon-2 favours them, but the audio still decides.

Languages

English only. Leaving the language empty, or on detect, still transcribes English. A project set to another language is refused: use Whisper (offline) for those.

en

Platforms

System	Download	Tested
Linux x86-64	341.8 MB	On real machines, and in clean Ubuntu 22.04, Debian 10, Fedora and Arch containers

Linux x86-64 only for now, on distributions from 2018 on (glibc 2.28 or newer). macOS and Windows builds are not ready yet.

Troubleshooting

unsupported_languagePhonon-2 knows only English. For other languages, use Whisper (offline).
model_missingThe model couldn't be downloaded or loaded. Check your internet connection for the first use, and the Models folder if you set one.
bad_paramsSubtitld couldn't hand the audio over. The message says why.
oomOut of memory. Close other programs, or transcribe a range instead of the whole video.
disk_fullNot enough disk space for the model or the decoded audio. Free some and try again.
internalSomething failed inside the add-on, often ffmpeg decoding the audio. The message has the details.

License

Add-on

Apache-2.0

Engine

Apache-2.0

fermion-research, by Fermion Research

Model

CC-BY-4.0

Phonon-2 weights

Fine for commercial work. The add-on and its engine are Apache-2.0. The Phonon-2 weights are CC-BY-4.0, which asks for credit to Fermion Research wherever the weights themselves are shared; the add-on's NOTICE file carries it. The subtitles you make are yours.

Versions

0.0.1

Offline English speech-to-text for Subtitld with Phonon-2 from Fermion Research (fermion-research 0.2.11).

Linux x86-64 for now. The archive runs on glibc 2.28 and newer, and was tested from the archive in clean Ubuntu 22.04, Debian 10, Fedora 44 and Arch Linux containers, with a real transcription in each. macOS and Windows builds are not done yet.

The model is not included: Phonon-2 (164 MB, CC-BY-4.0) downloads once on first use into Subtitld's shared model cache, checked against a sha256 pinned in fermion-research. English only; the Whisper (offline) add-on covers other languages. The engine runs on torch, so the download is large, about 340 MB.

The add-on is Apache-2.0. The archive contains THIRD_PARTY_LICENSES/ with the licences of everything frozen into it; INDEX.txt lists every shared library and the notice that covers it. The Phonon engine and transformers are Apache-2.0, torch and numpy are BSD-3-Clause, and the Python runtime and its libraries are under their own licences.

The archive also ships LGPL code, and its complete source is attached here:

libquadmath (LGPL-2.1-or-later), in two builds, linked by the GCC Fortran runtime of numpy's and scipy's OpenBLAS: phonon2-addon-0.0.1-libquadmath-source.tar.gz.
libsndfile (LGPL-2.1-or-later) from soundfile, with LAME and mpg123 (LGPL) and libFLAC, libogg, libvorbis and Opus (BSD) linked into it: phonon2-addon-0.0.1-libsndfile-source.tar.gz.

Each tarball's README says which build it is and how to rebuild it. The add-on's own source is phonon2-addon-0.0.1-source.tar.gz.

Phonon-2 (offline, English)