---
title: Voice Models Catalog
description: All voice models available via the Xantly Voice API, grouped by type and provider.
---

# Voice Models Catalog

Xantly exposes 33 voice models across 6 providers. Use the `stt_model` / `tts_model` parameters on the voice endpoints to pick a specific model, or let Xantly auto-route based on language and latency budget.

All models route through the same API key as chat, there is no separate voice subscription. Voice usage bills per-minute (provider pass-through) plus a per-minute platform fee. See [voice-agents guide](/docs/guides/voice-agents) for the billing breakdown.

---

## Speech-to-Text (STT), 10 models

Use with `POST /v1/voice/transcribe` or as `stt_model` on `POST /v1/voice/chat`.

| Slug | Provider | Pricing Unit | Notes |
| --- | --- | --- | --- |
| `openai/whisper-1` | OpenAI | per minute | Classic Whisper, 50+ languages |
| `openai/gpt-4o-transcribe` | OpenAI | per minute | Improved accuracy over Whisper |
| `openai/gpt-4o-mini-transcribe` | OpenAI | per minute | Cost-optimized GPT-4o transcribe |
| `deepgram/nova-2` | Deepgram | per minute | Default for English, low latency |
| `deepgram/nova-3` | Deepgram | per minute | Latest Deepgram model, top accuracy |
| `deepgram/flux` | Deepgram | per minute | Streaming-optimized |
| `groq/whisper-large-v3` | Groq | per hour | Whisper on Groq LPU (217x real-time) |
| `groq/whisper-large-v3-turbo` | Groq | per hour | Faster Whisper variant (cheaper) |
| `elevenlabs/scribe_v2` | ElevenLabs | per hour | Scribe, long-form transcription |

---

## Text-to-Speech (TTS), 13 models

Use with `POST /v1/voice/synthesize` or as `tts_model` on `POST /v1/voice/chat`.

| Slug | Provider | Pricing Unit | Notes |
| --- | --- | --- | --- |
| `openai/tts-1` | OpenAI | per 1M characters | Standard OpenAI TTS |
| `openai/tts-1-hd` | OpenAI | per 1M characters | Higher fidelity OpenAI TTS |
| `openai/gpt-4o-mini-tts` | OpenAI | per mtok (mixed) | Default, cheap, good quality |
| `deepgram/aura-1` | Deepgram | per 1M characters | Deepgram Aura-1 voices |
| `deepgram/aura-2` | Deepgram | per 1M characters | Deepgram Aura-2 voices (latest) |
| `elevenlabs/eleven_flash_v2` | ElevenLabs | per 1M characters | Fast, natural, cost-effective |
| `elevenlabs/eleven_flash_v2_5` | ElevenLabs | per 1M characters | Improved Flash V2.5 |
| `elevenlabs/eleven_multilingual_v2` | ElevenLabs | per 1M characters | Multilingual, high quality |
| `elevenlabs/eleven_v3` | ElevenLabs | per 1M characters | Latest ElevenLabs flagship |
| `google/gemini-2.5-flash-preview-tts` | Google | per mtok (mixed) | Gemini 2.5 Flash TTS |
| `google/gemini-2.5-pro-preview-tts` | Google | per mtok (mixed) | Gemini 2.5 Pro TTS |

---

## Realtime (bidirectional audio), 3 models

Use with `GET /v1/voice/realtime` (WebSocket upgrade).

| Slug | Provider | Notes |
| --- | --- | --- |
| `openai/gpt-4o-realtime-preview` | OpenAI | Full-duplex voice, tools, interruptions |
| `openai/gpt-4o-mini-realtime-preview` | OpenAI | Cheaper realtime variant |
| `google/gemini-2.0-flash-live` | Google | Gemini Live API bidirectional |

---

## Audio LLMs, 7 models

Text + audio input, text + audio output. Use via `POST /v1/chat/completions` with audio content parts, or via the voice endpoints.

| Slug | Provider (Route) | Notes |
| --- | --- | --- |
| `openai/gpt-4o-audio-preview` | OpenAI direct | Audio-aware GPT-4o |
| `openai/gpt-4o-mini-audio-preview` | OpenAI direct | Cheaper audio-aware variant |
| `openrouter/openai/gpt-audio` | OpenRouter → OpenAI | Routed via OpenRouter key pool |
| `openrouter/openai/gpt-audio-mini` | OpenRouter → OpenAI | Routed via OpenRouter key pool |
| `openrouter/openai/gpt-4o-audio-preview` | OpenRouter → OpenAI | Routed via OpenRouter key pool |
| `openrouter/mistralai/voxtral-small-24b` | OpenRouter → Mistral | Voxtral for audio understanding |
| `openrouter/xiaomi/mimo-v2-omni` | OpenRouter → Xiaomi | Omni audio-visual-text model |

---

## Music Generation, 2 models

Text → audio (music). Use via `POST /v1/chat/completions` with the model slug.

| Slug | Provider | Notes |
| --- | --- | --- |
| `openrouter/google/lyria-3-pro` | Google (via OpenRouter) | Lyria 3 Pro music generation |
| `openrouter/google/lyria-3-clip` | Google (via OpenRouter) | Lyria 3 Clip short music snippets |

---

## Speech-to-Speech, 1 model

Audio → audio transformation (voice conversion).

| Slug | Provider | Notes |
| --- | --- | --- |
| `elevenlabs/eleven_multilingual_sts_v2` | ElevenLabs | Voice conversion, multilingual |

---

## Auto-routing defaults

If you omit `stt_model` / `tts_model`, Xantly auto-routes:

- **STT**: English → `deepgram/nova-2`, other languages → `openai/whisper-1`
- **TTS**: `openai/gpt-4o-mini-tts` by default, or ElevenLabs/Deepgram/Groq when `provider` is set on the request

## Key rotation and fallback

All models that go through OpenRouter, NVIDIA, or Groq use round-robin key rotation across your configured API key pool (e.g. `GROQ_API_KEY_1..6`). On a per-key 402 / 429, requests automatically waterfall to the next key in the pool. See [reliability guide](/docs/guides/reliability) for details.

## Usage example

```bash
# Transcribe with Groq Whisper (cheapest STT)
curl -X POST https://api.xantly.com/v1/voice/transcribe \
  -H "Authorization: Bearer $XANTLY_API_KEY" \
  -F "audio=@input.wav" \
  -F "stt_model=groq/whisper-large-v3-turbo"

# Synthesize with ElevenLabs Flash (low-latency TTS)
curl -X POST https://api.xantly.com/v1/voice/synthesize \
  -H "Authorization: Bearer $XANTLY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Hello from Xantly",
    "model": "elevenlabs/eleven_flash_v2_5",
    "voice": "21m00Tcm4TlvDq8ikWAM",
    "output_format": "pcm_16000"
  }' --output hello.pcm

# Full pipeline with explicit STT + TTS picks
curl -X POST https://api.xantly.com/v1/voice/chat \
  -H "Authorization: Bearer $XANTLY_API_KEY" \
  -F "audio=@question.wav" \
  -F "stt_model=groq/whisper-large-v3-turbo" \
  -F "tts_model=deepgram/aura-2"
```