BlueTTS offline, voice clone, Hebrew + 6 languages

VoicesOfflineCPU onlyVoice cloningCommercial use OK

Dubs your subtitles in Hebrew, English, German, Spanish, Italian, Russian and Yiddish with BlueV3, by Maxim Melichov, in one of 15 voices or in each speaker's own voice cloned from about ten seconds of the video. It runs on your own computer's processor, three to five times faster than real time, with no graphics card.

Install it from Subtitld

Settings Add-ons BlueTTS (offline, voice clone, Hebrew + 6 languages) Install

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

Source on GitHub
Languages
7
Models
5
Voices
16
Add-on download
37–62 MB
RAM, at least
2.5 GB

What it does

Hebrew and Yiddish

BlueTTS was made for Hebrew first: it reads unpointed Hebrew text, numbers, dates and prices included, and English words inside a Hebrew line are read in English. It also speaks Yiddish, which few dubbing engines do.

The speaker's own voice

Subtitld cuts about ten seconds of each speaker from the video and sends them with every line. BlueTTS clones that voice in any of its seven languages. Or pick one of 15 voices.

Fast on any computer

On a laptop processor it speaks three to five times faster than real time. No graphics card is used.

Offline after the first line

Each part downloads once, the first time a line needs it. After that, nothing leaves your computer.

Before you install

What BlueTTS needs depends on the languages you dub:

  • Disk: 428 MB for the model, downloaded on the first line. Hebrew adds 1.2 GB (or 312 MB with the compact reading model), Yiddish 307 MB, and the speaker's own voice 136 MB, each the first time you use it.
  • Memory: about 700 MB while it speaks, about 2 GB once Hebrew is loaded.
  • Time: on an Intel Core i7-13700T, a second of speech takes about a fifth of a second to make, a third of a second in Hebrew. The first Hebrew line waits about 15 seconds more while the reading model loads.
  • Processor: any 64-bit Intel or AMD processor, or Apple Silicon. No graphics card is used.

Voices

In the Speakers tab, choose BlueTTS 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.

16 voices.

Voices
Women Noa, Maya, Lily, Tamar, Shira, Yael, Emma, Sara
Men Adam, Daniel, David, Jonathan, Ethan, Michael, Ariel

Every voice speaks all seven languages, each at its own natural pace in each language. They were made from readers of the LibriTTS-R audiobook corpus.

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.

BlueTTS listens to the first nine seconds or so of the sample, and copies the voice's timbre and pitch, not its delivery. The best sample is clean speech close to the microphone: noise and echo make the clone less like the speaker, and a phone recording gives a phone-sounding voice.

Models

The model and the reading models are not inside the add-on: each downloads into Subtitld's shared model cache the first time a line needs it, from Hugging Face at a pinned revision, each file checked. An interrupted download resumes where it stopped. Updating or removing the add-on keeps them.

Model DownloadGood for
BlueV3 v3.0.1 (ONNX) with the G2P data of en/de/es/it/ru Default bluev3
428 MB
Every line: the voices, and how English, German, Spanish, Italian and Russian are read
BlueV3 reference encoder (voice cloning) bluev3-reference-encoder
136 MB
The speaker's own voice, the first time you dub one
RenikudPlus Hebrew G2P renikudplus
1.2 GB
Hebrew, the first time you dub it (the default)
RenikudPlus Hebrew G2P, compact (int8) renikudplus-int8
312 MB
Hebrew, when you choose the compact reading model
ReNikud-yi Yiddish context model renikud-yi
307 MB
Yiddish, the first time you dub it

The add-on runs BlueTTS's own code on ONNX Runtime, without PyTorch. With the same voice, line and seed, it gives exactly what BlueTTS's own Python package gives.

Options

In Settings › Add-ons › BlueTTS (offline, voice clone, Hebrew + 6 languages) › Configure.

CPU threads
Default: 0

0 uses the computer's logical processors, at most 8. More is rarely faster; fewer keep the computer more responsive while it dubs.

Quality steps
Default: 8

How many passes the model makes over each line. 8 is BlueTTS's own setting. 4 is about 1.7 times as fast, with slightly lower quality.

Hebrew reading model
Default: accurate

How Hebrew text is turned into sounds. accurate is the full model, 1.2 GB, with about 2 GB of memory. compact is the same model made smaller, 312 MB: in our test it read about 1 word in 200 differently, more often wrongly.

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: about 10 seconds of one person speaking, clean, not a phone recording.

Languages

Hebrew, English, German, Spanish, Italian, Russian and Yiddish. A project in any other language gets a message that says so: pick another dubbing engine for it.

Spanish follows the project's region: Spain gets Castilian, any other region Latin American Spanish. Numbers, dates, prices and codes are read as words, except in Yiddish: write Yiddish numbers out as words, or the line is refused with a message that says so.

  • he
  • en
  • de
  • es
  • it
  • ru
  • yi

Platforms

System DownloadTested
Linux x86-64 61.7 MBDubbed with the real model on a real machine, through Subtitld's add-on manager, and in CI
macOS (Apple silicon) 43.8 MBDubbed with the real model in CI (Apple Silicon)
Windows x86-64 37.3 MBDubbed with the real model in CI

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 the models and dubs a line in every language.

Troubleshooting

  • network_unavailableA model couldn't be downloaded. Check your connection and try again: the download resumes.
  • disk_fullNot enough disk space for the model: 428 MB, and up to 1.7 GB more for Hebrew, Yiddish and the speaker's own voice.
  • unsupported_languageThe project's language is not one of the seven, 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, a Yiddish line has digits, 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: BlueTTS needs about 1 GB free, 2.5 GB with Hebrew.
  • internalSomething failed inside the add-on; the message has the details.

License

Add-on
Apache-2.0
Engine
MIT
BlueTTS, by Maxim Melichov; ONNX Runtime is MIT
Model
MIT, BigScience OpenRAIL-M
BlueV3 weights; the vocoder, fine-tuned from Supertone's Supertonic-3, is OpenRAIL-M, with use restrictions
Voices
CC-BY-4.0
Made from LibriTTS-R speakers
Reading models
Apache-2.0, MIT, CC BY-SA
RenikudPlus (Hebrew), polyg2p; the German and Italian lexicons and part of the Yiddish one are CC BY-SA

The add-on is Apache-2.0 and BlueTTS's code is MIT. The voice model is MIT too, except its vocoder, the part that turns the model's output into sound: it is fine-tuned from Supertone's Supertonic-3 and carries the BigScience OpenRAIL-M licence. Commercial work is allowed, but OpenRAIL-M forbids some uses of what it makes, among them breaking the law, deceiving people and impersonating someone without their consent. The 15 voices are CC-BY-4.0. The German and Italian lexicons, and part of the Yiddish one, are CC BY-SA; they download with the models and stay as their authors publish them.

The dubs you make are yours. A cloned voice sounds like a real person: clone voices you have the right to use, and say that a dub is synthetic where it matters.

Versions

  • 26.11.1

    Text-to-speech for Subtitld with BlueTTS (BlueV3 v3.0.1, by Maxim Melichov): Hebrew, English, German, Spanish, Italian, Russian and Yiddish at 44.1 kHz, in 15 stock voices or in the speaker's own voice cloned from the video.

    The archive holds the add-on, ONNX Runtime and the text front ends. The model (428 MB) downloads on the first line; Hebrew (1.2 GB, or 312 MB compact), Yiddish (307 MB) and cloning (136 MB) download when a line first needs them. Every file is pinned to a revision and checked by sha256. It runs on the CPU, several times faster than real time.

    The add-on is Apache-2.0; BlueTTS's code is MIT. The model's vocoder weights are under BigScience OpenRAIL-M, whose use restrictions apply to the speech it makes; the stock voices are CC-BY-4.0 (LibriTTS-R); the German, Italian and some Yiddish lexicons are CC BY-SA. Each archive carries THIRD_PARTY_LICENSES/ with the licence of everything frozen into it. The Linux archive carries libquadmath (LGPL-2.1, with numpy); its complete source is bluetts-<version>-libquadmath-source.tar.gz in this release.

Other Voices add-ons