Dubs your subtitles in 16 languages with KittenML's KittenTTS 2, in each speaker's own voice cloned from the video, or in one of 115 built-in voices. It runs on your own computer's processor, at about real time, with no graphics card.

What it does

The speaker's own voice

Subtitld cuts a few seconds of each speaker from the video and sends them with every line, with the words said in them. KittenTTS 2 clones that voice and speaks the translated line in it, so the dub sounds like the person on screen.

16 languages, 115 voices

English, Spanish, French, German, Italian, Portuguese, Russian, Chinese, Japanese, Korean, Arabic, Hindi, Turkish, Dutch, Polish and Indonesian, with 2 to 50 built-in voices each.

About real time, on a CPU

A second of speech takes 1.1 to 1.5 seconds to make on a laptop processor, with about 2.3 GB of memory. No graphics card is used.

Offline after the first line

The model downloads once, the first time you dub with it. After that, nothing leaves your computer.

A restrictive licence on the model. The add-on is Apache-2.0, but the KittenTTS 2 weights it downloads are KittenML's, under the Stellon Labs Community License. Commercial use, dubbing for pay included, is free only while you and your affiliates have less than US$1,000,000 in annual revenue and less than US$1,000,000 in total funding, and you must register with Stellon Labs. Use must also follow an Acceptable Use Policy that Stellon Labs does not publish, and the licence can be revoked. Read the licence before using it for paid work. If its terms don't suit you, VoxCPM2 and Sopro also clone voices, under Apache-2.0.

Before you install

Disk: about 1.7 GB for the model, downloaded on the first line, and 540 MB more for cloning, on the first cloned line.
Memory: about 2.3 GB while it speaks, so about 3 GB free.
Processor: on Intel and AMD, one with AVX2, FMA and F16C, which Intel processors have had since 2013 and AMD since 2015. Some Celeron, Pentium and Atom processors lack them, and there the first line fails with a message that says so. Every Apple Silicon Mac works.

Clone a voice only with its owner's consent.

Voices

In the Speakers tab, choose KittenTTS 2 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. Set the project's language before you dub.

116 voices in 16 languages.

ar 4
de 4
en 50
es 5
fr 5
hi 5
id 2
it 6
ja 4
ko 3
nl 3
pl 4
pt 6
ru 7
tr 3
zh 4

With Speaker's own voice, Subtitld cuts a sample of that speaker from the project's audio, 3 to 12 seconds, and sends it with the subtitle text spoken in it. KittenTTS continues from that sample, which comes closest to the speaker and is the steadiest. In our test, with English samples and Portuguese lines, the cloned dubs sounded nearly as close to their speakers as VoxCPM2's, and got fewer words wrong. 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.

Each built-in voice belongs to one language: it is a recording, and the model continues in its accent. So the Voice list shows the voices of your project's language. English has 50, among them KittenTTS 0.8's Bella, Jasper, Luna, Bruno, Rosie, Hugo, Kiki and Leo, and characters such as Marcus, a smooth radio DJ, or Victoria, a regal narrator. The other languages have 2 to 7 each. KittenML doesn't say where the recordings come from.

Now and then the model falls silent instead of speaking a line. The add-on notices and tries again, up to three times, so you don't have to.

Models

None of KittenML's files is inside the add-on: they download into Subtitld's shared model cache the first time you dub, from KittenML's Hugging Face repository at a pinned revision, each file checked. The reference encoder, which turns a sample into a voice, comes from the add-on's own release on the first cloned line. An interrupted download resumes where it stopped. Updating or removing the add-on keeps the models.

Model	Download	Good for
KittenTTS 2 (1.7B, ternary) Default kitten-tts-2	1.6 GB	Every voice and language, about real time on a CPU
Reference encoder, for cloning ref-encoder	517 MB	Only for the cloned voice, downloaded on its first line

The add-on runs KittenML's own C++ program for KittenTTS 2, which is about four times faster than KittenML's Python package and needs a third of its memory.

Options

In Settings › Add-ons › KittenTTS 2 (offline, voice clone, 16 languages) › Configure.

Delivery

Default: stable

Stable suits almost every project. Expressive is livelier, and a line comes out odd more often.

CPU threads

Default: 0

0 uses the performance cores of a processor that also has efficiency cores, else one thread per core, 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: 3 to 30 seconds of one person speaking.

Fallback reference transcript

Default: empty

The exact words spoken in the fallback clip. With them the clone is closer to the speaker and makes fewer mistakes.

Clone the timbre only

Default: off

Leaves the words of the speaker's clip out, and clones only the timbre. Try it if a dub keeps the speaker's accent, or when the subtitles paraphrase what is said.

Languages

English, Spanish, French, German, Italian, Portuguese, Russian, Chinese, Japanese, Korean, Arabic, Hindi, Turkish, Dutch, Polish and Indonesian. A project in any other language gets a message that says so: pick another dubbing engine for it.

English text is read the way it's meant: "the 3rd of May" becomes "the third of May", "$12.50" becomes "twelve dollars fifty". In the other languages, write numbers and symbols as words where it matters.

en
es
fr
de
it
pt
ru
zh
ja
ko
ar
hi
tr
nl
pl
id

Platforms

System	Download	Tested
Linux x86-64	136.1 MB	Dubbed with the real model, cloning included, on a real machine and in CI
macOS (Apple silicon)	109.2 MB	Dubbed with the real model, cloning included, in CI (Apple Silicon)
Windows x86-64	96.4 MB	Dubbed with the real model, cloning included, 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 model and dubs with it, in a cloned voice too. On Linux it needs glibc 2.35 or newer.

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, about 1.7 GB, or 2.2 GB with cloning.
unsupported_languageThe project's language is not one of the 16, 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.
internalSomething failed inside the add-on; the message has the details. If it says the processor lacks AVX2, FMA and F16C, this computer can't run KittenTTS 2. If it says the line kept producing silence, try another voice or reword the line.

License

Add-on

Apache-2.0

Engine

MIT, Apache-2.0

kitten-tts-2-cpp, by KittenML; LibTorch is BSD-3-Clause

Reference encoder

MIT

From chatterbox-turbo (ResembleAI) and pyannote

Model

Stellon Labs Community License

KittenTTS 2 weights; commercial use free only below US$1M in revenue or funding

The add-on is Apache-2.0. KittenML's runtime is MIT and Apache-2.0, LibTorch is BSD-3-Clause, and the reference encoder's parts are MIT. The KittenTTS 2 weights are under the Stellon Labs Community License, which you accept when the add-on downloads them: free for research and personal use, and for commercial use only below US$1,000,000 in annual revenue and in total funding, with registration with Stellon Labs. The weights and what they make may not be used to train other models. The dubs you make are yours. Powered by KittenML.

KittenTTS 2 (offline, voice clone, 16 languages)