# Voices and languages

> The voices you can speak with, how to list them, and the twelve languages the model speaks.

## Voices

Pass a voice’s name as `voice`, in any case: `Monika` and `monika` are the same voice.

| voice | Made for | Gender | Description |
| --- | --- | --- | --- |
| Bunty | Hindi | Male | Young male voice, friendly, casual |
| Monika | Hindi | Female | Young female voice, warm, helpful |
| Umananda | Assamese | Male | Adult male voice, steady, friendly |
| Rimjhim | Bengali | Female | Young female voice, clear, expressive |
| Muskan | Gujarati | Female | Adult female voice, warm, sweet |
| Srivatsa | Kannada | Male | Adult male voice, casual, natural |
| Mahendran | Malayalam | Male | Young male voice, relatable, natural |
| Sunaina | Malayalam | Female | Adult female voice, warm, clear |
| Amogh | Marathi | Male | Young male voice, casual, engaging |
| Kalyani | Marathi | Female | Young female voice, warm, gentle |
| Swastik | Odia | Male | Adult male voice, casual, warm |
| Raqib | Punjabi | Male | Young male voice, warm, conversational |
| Vennila | Tamil | Female | Young female voice, warm, easygoing |
| Bhaavana | Telugu | Female | Young female voice, natural, warm |

Each voice is made from a recording in its own language, and can speak the other eleven too. Hear them all in the dashboard’s [playground](https://app.vaakini.com/app/playground).

Your org can also have voices of its own, each from a short recording: [custom voices](https://docs.vaakini.com/custom-voices.md), spoken by their id (`cv_…`).

## Listing voices

`GET /v1/voices` answers the voices your org can speak with: ours, then your custom voices.

**cURL**

```shell
curl https://api.vaakini.com/v1/voices \
  -H "Authorization: Bearer $VAAKINI_API_KEY"
```

**Python**

```python
from vaakini import Vaakini

client = Vaakini()  # reads VAAKINI_API_KEY

for voice in client.voices.list():
    print(voice.id, voice.category, voice.gender, voice.languages)

model = client.models.list()[0]
print(model.id, model.languages, model.formats)
```

```json
{
  "voices": [
    {
      "id": "Monika",
      "name": "Monika",
      "category": "builtin",
      "gender": "female",
      "description": "Young female voice, warm, helpful",
      "languages": ["hi"]
    },
    {
      "id": "cv_t76u7a3um3xayk4b",
      "name": "Priya (support)",
      "category": "cloned",
      "gender": "female",
      "description": null,
      "languages": ["hi"]
    }
  ]
}
```

`id` is what `voice` takes. `category` is `builtin` for one of ours and `cloned` for a custom voice. `languages` is the language the voice is made for. The list is the one to trust: a voice added since this page was written is in it. `GET /v1/models`, which needs no key, lists our voices’ names alone.

## Languages

`language` says which language the text is in. It is optional.

| language | Language |  |
| --- | --- | --- |
| as | Assamese | অসমীয়া |
| bn | Bengali | বাংলা |
| en | English | English |
| gu | Gujarati | ગુજરાતી |
| hi | Hindi | हिन्दी |
| kn | Kannada | ಕನ್ನಡ |
| ml | Malayalam | മലയാളം |
| mr | Marathi | मराठी |
| or | Odia | ଓଡ଼ିଆ |
| pa | Punjabi | ਪੰਜਾਬੀ |
| ta | Tamil | தமிழ் |
| te | Telugu | తెలుగు |

Left out, the model picks the language from the script of each sentence. Name it when you know it: with [normalization](https://docs.vaakini.com/options.md#normalize) on, numbers, dates and amounts are read aloud in the request’s language.

Any voice can be combined with any language.

## Write each word in its own script

> A word is pronounced most reliably when it is written in the script of its language: a Hindi word in Devanagari, a Tamil word in Tamil script, an English word in Latin letters. The same Hindi word spelled in Latin letters (`dhanyavaad` for `धन्यवाद`) is more likely to be mispronounced.
>
> If an LLM writes your agent’s replies, say so in its system prompt, for example: “Write every word in the script of its own language: Hindi words in Devanagari, English words in English letters. Never write a Hindi word in English letters.”
