Speech

Synthesise speech from text. The endpoint returns the raw audio bytes as the response body - no JSON envelope.

POST /v1/audio/speech

Request body

Example

curl https://api.aivene.com/v1/audio/speech \
  -H "Authorization: Bearer $AIVENE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tts-1",
    "voice": "alloy",
    "input": "The quick brown fox jumps over the lazy dog."
  }' \
  --output speech.mp3

With the OpenAI SDK:

import { writeFile } from 'node:fs/promises';

const res = await client.audio.speech.create({
  model: 'tts-1',
  voice: 'alloy',
  input: 'Hello from Aivene.'
});

const buffer = Buffer.from(await res.arrayBuffer());
await writeFile('speech.mp3', buffer);

Picking a model

If you need to play audio before generation finishes, use tts-1 and stream the bytes to your audio sink as they arrive.

Picking a voice

The voice ids above are deliberately generic. Their characteristics:

Run a quick sample with each before committing - voice perception is subjective.

Format and quality

speed affects the wall-clock duration but not the pitch.

Streaming playback

The endpoint streams audio bytes as they are generated. With fetch:

const res = await fetch('https://api.aivene.com/v1/audio/speech', {
  method: 'POST',
  headers: {
    Authorization: `Bearer ${process.env.AIVENE_API_KEY}`,
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    model: 'tts-1',
    voice: 'nova',
    input: 'Streaming this sentence as it arrives.'
  })
});

const reader = res.body?.getReader();
while (reader) {
  const { done, value } = await reader.read();
  if (done) break;
  // Pipe `value` (a chunk of audio bytes) into your audio sink.
}

In the browser, feed the chunks into a MediaSource to play before the whole clip lands.

Cost notes

TTS bills per character of input text, regardless of model or voice. The Console usage page shows the per-request character count.

Errors

Error bodies are always JSON, even though success bodies are audio. Check Content-Type on the response: audio/* for success, application/json for errors.

Next steps

• Audio - go the other direction with transcription.
• Chat Completions reference - generate the text first, then speak it.