Which voices can I use?

The reference examples use onyx and shimmer. The voices available depend on the model you pick — see the Audio Speech reference.

Which audio formats are supported?

mp3 by default, plus opus, aac, flac, wav, and pcm — set response_format to choose.

What if my text is longer than 4,096 characters?

The request is rejected with HTTP 400. Split the text into chunks under the limit, send one request per chunk, and play or join the results in order.

Do I need a separate key for speech?

No. One virtual key serves every modality, and its budget covers all of them.

Generate Speech

Turn text into spoken audio through nRouter with one POST to /v1/audio/speech, using the same key, budgets, and request logs as your chat traffic.

Turn text into spoken audio with one POST to /v1/audio/speech. Send the text and a voice; the response is the audio file. Speech runs through the same key, budgets, and request logs as your chat traffic.

Send the request

curl -X POST https://api.nrouter.ai/v1/audio/speech \
  -H "Authorization: Bearer $NROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tts-1",
    "input": "Your order has shipped.",
    "voice": "onyx",
    "response_format": "mp3"
  }' \
  --output speech.mp3

response_format defaults to mp3; opus, aac, flac, wav, and pcm are also supported.

Voices, speed, and length

The voices are alloy, echo, fable, onyx, nova, and shimmer. The speed field sets the playback rate from 0.25 (slowest) to 4.0 (fastest) and defaults to 1.0. One request takes up to 4,096 characters of input — longer text is rejected with HTTP 400, so split long passages into chunks and stitch the resulting audio files in order.

Choose the format for where the audio plays, not just for file size:

mp3 — universal playback in browsers and mobile apps (the default)
opus — low-bitrate streaming, VoIP, and WebRTC pipelines
aac — playback across iOS and Apple ecosystems
flac — lossless audio for archival and post-processing
wav — uncompressed audio for players that should not spend time decoding
pcm — raw 24 kHz 16-bit samples for digital signal processing

What you get back

The response is the audio file itself, in the format you requested — save it, stream it, or hand it to your player. The call appears in your request logs with its exact cost, like any other request. The full field list lives in the Audio Speech API reference, and errors use the same shape as everywhere — see Errors & Rate Limits.

FAQ

Next steps

Audio Speech API — The full request and response reference
Generate Images — One-POST image generation
Errors & Rate Limits — Every status, code, and retry rule