Ming-3.0-tts (offline, voice clone, 7 languages)

What it does

Before you install

Ming-3.0-tts is a large model, about 1.5 billion parameters with its audio codec. It asks a lot of your computer:

• Disk: about 3.4 GB, most of it the 2.99 GB model, downloaded on the first line you dub.
• Memory: about 5 GB free. With less it still runs, but far slower.
• Time: on a CPU each second of speech takes about four to five seconds to make, on an Intel Core i7-13700T. A ten-minute video can take about three quarters of an hour.

For quick dubs on modest computers, Piper or Kokoro are much lighter.

Voices

In the Speakers tab, choose Ming-3.0-tts 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.

12 voices.

The cloned voice

With Speaker's own voice, Subtitld cuts a sample of that speaker from the project's audio, from one of their subtitles of 3 to 12 seconds, 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.

Subtitld sends the subtitle text with the sample, and Ming-3.0-tts carries on from it, which comes closest to the speaker, delivery included. If a dub into another language keeps the speaker's accent, turn on Clone the timbre only.

Built-in voices

Emma (American), George and Charlotte (British), Ling Xiaorou and Ling Kai (Chinese) are speakers the model learned. Each sounds the same on every line, and each speaks all seven languages.

Described voices

Ming-3.0-tts makes up a new voice every time it reads a description. So the add-on makes each speaker's voice once, and every line of that speaker reuses it: the voice stays the same through the video, and two speakers given the same description still sound like two people.

Emotion, speed, pitch and volume

These are settings of the add-on, so they apply to every speaker dubbed with a cloned or described voice. They are the model's own controls, which take the sound of the speaker's voice but not their way of speaking, and they work one at a time: an emotion, when set, overrides speed, pitch and volume. Built-in voices ignore them.

In Chinese or English lines, the model also makes sounds written into the text: [Laughter], [Sigh], [Breathing], [Cough], [Uhm] and a few more.

Models

There is one model, Ming-3.0-tts-0.5B. It is not inside the add-on: it 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 model.

The first line also loads the model, so it takes longer than the ones after it. Cancel stops a line between two steps, or stops the download.

Options

In Settings › Add-ons › Ming-3.0-tts (offline, voice clone, 7 languages) › Configure.

Languages

Chinese, English, Japanese, Korean, German, French and Spanish. Every voice speaks every one of them.

• zh
• en
• ja
• ko
• de
• fr
• es

Platforms

Linux x86-64, macOS on Apple Silicon and Windows x86-64. Each release is built on all three, and CI checks that everything needed to speak is in the build. CI doesn't run the model itself.

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. It needs about 3.4 GB free.
• oomOut of memory. Close other programs; Ming-3.0-tts needs about 5 GB free.
• bad_paramsThe line has no text, the cloned voice got no sample, or Voice you describe has no description. The message says which: open the video, set a Fallback reference clip, or write a Voice description.
• unsupported_voiceThe project names a voice this version doesn't have. Pick the voice again in the Speakers tab.
• internalSomething failed inside the add-on; the message has the details. If you changed Inference device, set it back to Automatic.

License

Fine for commercial work: the add-on is Apache-2.0, and Ming-3.0-tts's code and the Ming-3.0-tts-0.5B weights are MIT. 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.