Speech
Turn text into spoken audio.
POST /v1/audio/speech submits a text to speech job. It answers 202 with a job, not an audio
file; see How media jobs work for polling and output.
curl https://api.opinference.com/v1/audio/speech \
-H "Authorization: Bearer $OP_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "hexgrad/kokoro-82m",
"prompt": "Welcome to OP Inference.",
"voice": "af_heart"
}'The text goes in prompt, not input as in the OpenAI speech API.
Inputs
| Input | Meaning |
|---|---|
prompt | The text to speak |
voice | A voice name. Each model has its own list |
language | The language, where the model asks for one |
speed | Speaking rate, where the model offers it |
The model's page at opinference.com/models lists its voices and limits.
Output
One audio file in output[], such as audio/wav. Speech is usually priced per character of
prompt, and usage.units.characters counts them.