Chatterbox offline, voice clone, 20 languages

VoicesOfflineCPU onlyVoice cloningCommercial use OK

Dubs your subtitles in 20 languages with Resemble AI's Chatterbox, in each speaker's own voice cloned from a few seconds of the video, or in one of 8 preset voices. A cloned voice speaks every language, whatever language the clip is in. It runs on your own computer's processor at about real time, with no graphics card.

Install it from Subtitld

Settings Add-ons Chatterbox (offline, voice clone, 20 languages) Install

Subtitld downloads it, checks its SHA-256, and keeps it updated.

Source on GitHub
Languages
20
Models
2
Voices
9
Add-on download
39–62 MB
RAM, at least
2 GB

Listen

What it does

The speaker's own voice, in another language

Subtitld cuts a few seconds of each speaker from the video and sends them with every line. Chatterbox clones that voice in any of its 20 languages, so a Portuguese actor can be dubbed in German and still sound like themselves.

Two models

Multilingual speaks all 20 languages. Turbo speaks English only, sounds closer to the speaker, needs less memory and reads tags such as [laugh] and [sigh] written in the subtitle.

On any computer

It runs on the processor at about real time, with no graphics card: a 10-second line takes about 10 seconds.

Free for commercial work

The add-on is Apache-2.0, Chatterbox and its models are MIT, and the preset voices are LibriVox readers, in the public domain.

Before you install

  • Disk: Multilingual downloads 1.3 GB on its first line, Turbo 725 MB.
  • Memory: about 1.7 GB while it speaks with Multilingual, about 1 GB with Turbo.
  • Processor: any 64-bit Intel or AMD processor, or Apple Silicon. No graphics card is used.

Clone a voice only with its owner's consent. Chatterbox's own builds mark their audio with an inaudible watermark; this add-on does not, so the responsibility for how a cloned voice is used is yours.

Voices

In the Speakers tab, choose Chatterbox as a speaker's Dubbing engine, then pick a Voice. Turn on Fit dub to subtitle duration for the speaker to stretch or squeeze each line to its subtitle.

9 voices.

With Speaker's own voice, Subtitld cuts a sample of that speaker from the project's audio, from one of their subtitles or several short ones joined. If you have separated music from voice for this video, the sample comes from the voice alone. The Fallback reference clip in the add-on's settings is used only when no sample can be cut. The best sample is 5 to 20 seconds of the speaker alone, with no music or other voices; the add-on uses its first 10 seconds.

Preset voices
Women English, Portuguese, French, German
Men English, Portuguese, French, German

Each preset is a LibriVox volunteer reading a book in that language, the same readers as in the Sopro add-on. Every preset speaks all 20 languages; its own language only gives it an accent, so pick the closest one to your project's language for the most natural dub.

Models

Neither model is inside the add-on: the one you use downloads into Subtitld's shared model cache the first time you dub, from Hugging Face at a pinned revision, each file checked. An interrupted download resumes where it stopped. Updating or removing the add-on keeps the models.

ModelLanguage DownloadGood for
Chatterbox Multilingual (0.5B, 4-bit ONNX) Default multilingual
1.3 GB
All 20 languages; the default
Chatterbox Turbo (350M, 4-bit ONNX, English) turboen
725 MB
English only: closer to the speaker, less memory, and reads [laugh], [sigh] and [cough]

The add-on runs ONNX exports of Chatterbox, without PyTorch. In our test of 10 sentences in English, Portuguese, French, German, Spanish, Italian, Russian and Polish, each cloned from two voices, Multilingual got 2.4% of the words wrong and sounded close to the speaker; Turbo got none wrong in English and sounded closer still.

Options

In Settings › Add-ons › Chatterbox (offline, voice clone, 20 languages) › Configure.

Model
Default: multilingual

Turbo speaks English only. With Turbo picked, lines in other languages are spoken by Multilingual. Only one model is in memory at a time.

Expressiveness
Default: 0.5

Multilingual only: how much emotion the voice puts in. 0.5 is Chatterbox's own default; higher is more dramatic and tends to speak faster.

CPU threads
Default: 0

0 uses half of the computer's logical processors, at most 8. Fewer threads keep the computer more responsive while it dubs.

Fallback reference clip
Default: empty

A clip for the cloned voice when Subtitld can't cut one from the video, for example with no video open: 5 to 20 seconds of one person speaking.

Languages

Arabic, Danish, Dutch, English, Finnish, French, German, Greek, Hindi, Italian, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish and Turkish. A project in any other language gets a message that says so: pick another dubbing engine for it.

Chatterbox also has Hebrew, Japanese and Chinese, which the add-on leaves out: Hebrew needs a second model to add its vowel points, Japanese needs a converter whose licence doesn't allow bundling it, and Chinese came out garbled in our tests.

Write numbers and symbols as words where it matters. A line is spoken in one go, up to 40 seconds of speech.

  • ar
  • da
  • de
  • el
  • en
  • es
  • fi
  • fr
  • hi
  • it
  • ko
  • ms
  • nl
  • no
  • pl
  • pt
  • ru
  • sv
  • sw
  • tr

Platforms

System DownloadTested
Linux x86-64 62.5 MBDubbed with both models on a real machine, and in Subtitld itself
macOS (Apple silicon) 43.7 MBIn CI, the installed add-on downloads a model and dubs with it (Apple Silicon)
Windows x86-64 38.7 MBIn CI, the installed add-on downloads a model and dubs with it

Linux x86-64, macOS on Apple Silicon and Windows x86-64. Each release is built on all three, and on each one CI installs the add-on as Subtitld does, downloads a model and dubs with it.

Troubleshooting

  • network_unavailableThe model couldn't be downloaded. Check your connection and try again: the download resumes.
  • disk_fullNot enough disk space for the model: 1.3 GB for Multilingual, 725 MB for Turbo.
  • unsupported_languageThe project's language is not one of the 20, or not set. Set it in the project, or pick another dubbing engine.
  • unsupported_voiceThe project names a voice this version doesn't have. Pick the voice again in the Speakers tab.
  • bad_paramsThe line has no text, or the cloned voice got no sample: open the video, or set a Fallback reference clip. The message says which.
  • out_of_memoryThe computer ran out of memory. Close other programs, or pick Turbo for English: it needs about 1 GB.
  • internalSomething failed inside the add-on; the message has the details.

License

Add-on
Apache-2.0
Engine
MIT
Chatterbox, by Resemble AI; ONNX Runtime is MIT
Model
MIT
Chatterbox Multilingual and Turbo weights, in their ONNX exports
Preset voices
Public domain, CC-BY-4.0
LibriVox readers; the English ones through LibriTTS-R

Fine for commercial work: the add-on is Apache-2.0, and Chatterbox and both models are MIT. The preset voices are LibriVox recordings, in the public domain; the English ones come through LibriTTS-R, under CC-BY-4.0, and the add-on credits each one. The dubs you make are yours.

Versions

  • 26.10.1

    Voice-cloning text-to-speech for Subtitld with Resemble AI's Chatterbox: 20 languages at 24 kHz with Chatterbox Multilingual, or English with Chatterbox Turbo, in the speaker's own voice cloned from the video, or in a preset voice.

    The archive holds the add-on, ONNX Runtime and the preset voice clips. A model (Multilingual 1.3 GB, Turbo 0.7 GB) downloads on its first line, at a pinned revision, checked by sha256. It runs on the CPU at about real time.

    The add-on is Apache-2.0; the Chatterbox models are MIT. Each archive carries THIRD_PARTY_LICENSES/ with the licence of everything frozen into it, and the preset clips are credited in bin/_internal/chatterbox_addon/presets/README.md. The Linux archive carries libquadmath (LGPL-2.1, with numpy); its complete source is chatterbox-<version>-libquadmath-source.tar.gz in this release.

Other Voices add-ons