> For the complete documentation index, see [llms.txt](https://documentation.proto.cx/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://documentation.proto.cx/docs/developers/local-languages/language-training.md).

# Language training

### Text support vs voice support

| Type              | What it enables                                                                          | Data required          |
| ----------------- | ---------------------------------------------------------------------------------------- | ---------------------- |
| **Text support**  | Your AI Agent recognises and responds to written messages in the target language         | Text data only         |
| **Voice support** | Your AI Agent understands spoken messages and responds with voice in the target language | Text data + voice data |

All new languages require text data preparation. Voice data is only required if your AI Agent will support voice channels.

***

### Before you begin

Confirm the following with your Proto contact before starting any data work:

| Information     | Description                                                                                                                                                                 | Example                                                          |
| --------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------- |
| Target language | The language being added. If the language has regional variants, specify which one – regional variants can differ significantly in vocabulary, spelling, and pronunciation. | English (UK) vs English (US); French (France) vs French (Canada) |
| Use case        | The primary purpose of your AI Agent                                                                                                                                        | Complaint handling, customer FAQs, sales support                 |
| Channels        | Where the AI Agent will be deployed                                                                                                                                         | WhatsApp, Webchat, voice IVR                                     |
| Timeline        | Your expected go-live date                                                                                                                                                  | —                                                                |

{% hint style="info" %}
Why specifying the variant matters: Many languages have regional variants that differ in vocabulary, spelling, and pronunciation. English (UK) and English (US) are a familiar example – the same logic applies across many languages worldwide. Specifying the correct variant from the start ensures the AI Agent is trained on the right language for your audience.
{% endhint %}

***

### Guides

Follow the relevant guide or guides based on your requirements:

{% content-ref url="/pages/LffWZfYXvAgtNRag4bLY" %}
[Voice data collection](/docs/developers/local-languages/language-training/voice-data-collection.md)
{% endcontent-ref %}

{% content-ref url="/pages/dCS14f5rFaYHkrr9eCp6" %}
[Text data preparation](/docs/developers/local-languages/language-training/text-data-preparation.md)
{% endcontent-ref %}

* **Text data preparation** – required for all new languages
* **Voice data collection** – required for voice-enabled AI Agents only

***

### Common questions

**Do I need voice data if my AI Agent is text-only?**

No. Voice data is only required if your AI Agent will understand spoken messages or respond out loud. Text-only deployments only require the text data preparation guide.

**Can my translator use an AI translation tool to speed up the work?**

For widely spoken languages such as French, Spanish, or Arabic, AI translation tools can help with speed. For low-resource languages – languages with limited digital resources, such as many local African or Southeast Asian languages – reliable translation tools are unlikely to be available. In those cases, a human native speaker is required for all translation work.

**Can existing recordings be used for voice training?**

Yes. If you already have audio in the target language with matching transcripts, follow Option A in the voice data collection guide. Contact your Proto contact first to confirm the files meet the format requirements.

**What if my language is not widely supported by standard AI tools?**

Low-resource languages require more manual work throughout. All translation must be done by a human native speaker, and voice training requires the full recording process. Your Proto contact can advise on what to expect for your specific language.