Text to speech: formats, languages and streaming
The three ways to ask for speech, the formats you can get, and the options.
- Who can do this
- Everyone
- Reading time
- 1 min read
| Endpoint | You get |
|---|---|
| POST /v1/text-to-speech/{voice_id} | The audio |
| POST /v1/text-to-speech/{voice_id}/with-timestamps | The audio as base64, with the timing of every character |
| POST /v1/text-to-speech/{voice_id}/stream | The audio a sentence at a time, as soon as each is ready |
Formats
Set output_format in the query: MP3 at several bit rates (the default is mp3_44100_128), WAV, raw PCM, Opus, and ulaw_8000 and alaw_8000 for telephone lines.
Language
Without language_code, each sentence is read in the language it's mostly written in. Send en or sw to read the whole text in one language, which helps with short texts that could be either.
Getting the same audio again
Send a seed and the same text, voice and settings give the same audio. Navoras keeps it, and a repeat from the same workspace is answered from what was kept, for nothing: its character-cost header says 0.
Headers worth reading
request-id: quote it when you write to us about a request.character-cost: the credits the request cost.history-item-id: where to find it in history, when logging is on.
Still stuck?
A person at Navoras reads every message and replies by email. Name your workspace, and the request ID if it's about one.