## ❓ Finetuning for New Languages — Is It Supported?

[humair-m]

❓ Finetuning for New Languages — Is It Supported?

Hi team ,

I wanted to ask whether finetuning is currently supported for adding or improving a new language in this project.

My questions:

• Can the model be finetuned on a custom dataset for a language that is not currently supported?
• If yes, could you please share documentation or instructions (data format, scripts, configs, training pipeline)?
• If no, is this feature planned for future releases?

Thanks for the great work!
Looking forward to your guidance.

[dosubot]

Officially, OpenVoice does not provide built-in scripts or documentation for finetuning the model on new languages. However, the community has created a finetuning script with detailed instructions, available at OpenVoice_ONNX. This script covers directory setup, audio requirements (high-quality, noise-free .wav files, 2–20 seconds), finetuning commands, and model testing.

OpenVoice supports any language as long as you have a base speaker TTS model for that language—the tone color converter is already trained by the OpenVoice team. If your target language is not supported by an existing TTS model, you’ll need to train or obtain one separately; OpenVoice cannot synthesize output in a language without a base TTS model, even if you finetune the tone color converter itself (source).

There is no public roadmap or official announcement about native finetuning support for new languages in future releases. For now, community scripts are your best option.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[humair-m]

like for urdu @dosu

[dosubot]

OpenVoice can support Urdu, but only if you have a base speaker TTS model for Urdu. The core OpenVoice model (the tone color converter) is language-agnostic, but the actual speech synthesis depends on the base TTS model. Out of the box, OpenVoice only includes base models for English and Chinese, so Urdu is not natively supported.

To add Urdu, you would need to train or obtain a high-quality Urdu TTS model and integrate it as the base speaker in the OpenVoice pipeline. The community finetuning scripts (like OpenVoice_ONNX) can help with voice cloning, but they still require a working Urdu TTS model as the foundation. There are no official or community scripts or documentation specific to Urdu in this repo right now. For more details on how language support works and integration steps, see the official Q&A and usage docs.

If you can source or train an Urdu TTS model, you can plug it into OpenVoice by following the same process as for other languages. The main challenge is finding or building a good Urdu TTS model—once you have that, OpenVoice can handle the rest.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[humair-m]

ok , is any train.py so i train from scratch @dosu

[dosubot]

There isn’t a train.py or official script in the OpenVoice repo for training the tone color converter from scratch. Training from scratch is only supported for the TTS component (MeloTTS), which is a separate package with its own train.py and instructions—see the MeloTTS repo: https://github.com/myshell-ai/MeloTTS reference.

For OpenVoice itself, you can only finetune the tone color converter using a community script: OpenVoice_ONNX reference. If you want to support a new language, the workflow is: train or obtain a TTS model for that language (using MeloTTS or another TTS system), use OpenVoice for tone color conversion, and optionally finetune the tone color converter with the community script if needed.

OpenVoice and MeloTTS are architecturally separate—OpenVoice handles timbre transfer, while MeloTTS does the actual speech synthesis.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}