OpenRouter for Audio
One endpoint for every TTS, STT, and realtime voice model. Route by latency, region, provider, cost, or capability.
Keep the voice. Switch the model.
Your users know the voice they hear every day. Bring that voice to another provider when you need lower latency, better quality, a different region, or a different model family.
Send us a source provider and voice, choose a destination provider, and Allmodels returns a portable voice slug like elevenlabs/edward that can be used on compatible routes such as cartesia/sonic.
POST /v1/voices/port
{
"source": {
"provider": "elevenlabs",
"voice": "edward"
},
"destination": {
"provider": "cartesia"
}
}
{
"voice": "elevenlabs/edward",
"provider": "cartesia",
"status": "ready"
}One API call. Advanced audio features.
Ask for the capability you need. Allmodels maps the request to the provider-specific schema, stream, and model route that can satisfy it.
Voice cloning
Use one clone ID across providers. We translate provider-specific voice creation, voice IDs, and generation parameters.
Multi-voice
Send speaker arrays for dialogue, dubbing, podcasts, and agent handoffs without reimplementing each provider's dialogue API.
Multilingual STT/TTS
Route by language, accent, diarization, translation, timestamps, and model-level capability constraints.
Output stream you want
Request WebSocket, SSE, chunked HTTP, file response, PCM16, Opus, MP3, WAV, mu-law, verbose JSON, or word timestamps.
Region-aware routing
Default to the closest Allmodels edge and closest provider server, or pin provider.region directly.
Provider pass-through
One account, one usage surface, and routing controls inspired by OpenRouter's provider selection model.
Live stats
Mocked for the prototype, designed to become the public routing telemetry surface.
Last 60s - TTFB per provider
What we measure
TTFB, first audio, P50/P95 by provider, model, and region; error rates; fallback time; and stream format compatibility.