# Soniox STT

> Use Soniox real-time speech-to-text for multilingual and Indian-language voice agents, with language identification and optional speaker labels.

Source: https://edesy.in/docs/voice-agent/stt/soniox

Soniox is a real-time speech-to-text provider on the default Edesy runtime. It is a strong choice for Indian languages such as Malayalam, Tamil and Bengali, and it can identify the spoken language per utterance. For Hinglish calls where English words must stay in Latin script, compare it with [Sarvam](https://edesy.in/docs/voice-agent/stt/sarvam) in codemix mode.

## At a glance

| Item | Value |
|------|-------|
| Provider ID | `soniox` (the dashboard also stores `soniox_stt`) |
| Runtime | Default runtime |
| Default model | `stt-rt-v5` (`stt-rt-v4` and `stt-rt-preview` are served as the same model) |
| Languages in the dashboard | English (India, US, UK), Hindi, Tamil, Telugu, Kannada, Malayalam, Bengali, Gujarati, Marathi, Punjabi, Odia, Assamese |
| Language identification | On by default |
| Speaker labels | Optional |
| Your own key (BYOK) | Supported |

## When to use Soniox

- Calls in South Indian or eastern languages where transcript accuracy matters most.
- Callers who switch between languages and you want the model to detect the language itself.
- You already use Soniox elsewhere and want one vendor for [STT and TTS](https://edesy.in/docs/voice-agent/tts/soniox).

## Set it up

1. Open the agent and go to **Voice → Speech-to-text**.
2. Choose **Soniox**.
3. Keep the model on **Realtime v5**.
4. Leave **Strict Language Hints** off for phone callers who code-switch.
5. Run a test call and read the transcript in call history.

```json
{
  "sttProvider": "soniox",
  "sttConfig": {
    "model": "stt-rt-v5",
    "enable_language_identification": true,
    "language_hints_strict": false
  }
}
```

## Settings

| Setting | Default | What it does |
|---------|---------|--------------|
| `enable_language_identification` | `true` | Detects the language of each utterance |
| `language_hints_strict` | `false` | When on, only the hinted languages are transcribed. Keep it off for callers who mix Hindi, English and a regional language |
| `enable_speaker_diarization` | `false` | Adds speaker labels to the transcript |

## FAQs

**Does Soniox keep English words in English script?**
Not always. For Hinglish, test [Sarvam with codemix](https://edesy.in/docs/voice-agent/stt/sarvam) as well and keep the provider that gives cleaner transcripts for your callers.

**Is there an add-on charge?**
The dashboard catalog shows no STT add-on for Soniox. See [provider costs](https://edesy.in/docs/voice-agent/billing/provider-costs).

**Can I use my own Soniox key?**
Yes, through [BYOK](https://edesy.in/docs/voice-agent/workspace/byok).

**Which model should I pick?**
`stt-rt-v5`. The older names are aliases of the same model and stay listed only so existing agents show a selected value.

## Related

- [STT providers overview](https://edesy.in/docs/voice-agent/stt)
- [Soniox TTS](https://edesy.in/docs/voice-agent/tts/soniox)
- [Turn detection](https://edesy.in/docs/voice-agent/pipeline/turn-detection)