Speech Synthesis (TTS)

Convert text to speech using AI models.

Basic Usage

const response = await ai.speech.get({
  text: 'Hello! This is a text-to-speech example.',
  voice: 'alloy',
});

// Save to file
import fs from 'fs';
import { Readable } from 'stream';

const fileStream = fs.createWriteStream('output.mp3');
Readable.fromWeb(response.audio).pipe(fileStream);

Request Options

interface SpeechRequest {
  text: string;                 // text to speak
  voice?: string;               // voice name or custom voice id
  speed?: number;               // speaking speed (0.25 - 4.0)
  responseFormat?: AudioFormat; // 'mp3' | 'opus' | 'aac' | 'flac' | 'wav' | 'pcm'
  instructions?: string;        // style instructions (if supported)
}

Available Voices

Voices are provider-specific. The voice is passed through as given, so any built-in voice name or a custom voice id works.

OpenAI: built-in voices such as alloy (the default), marin, cedar, coral, ash, or a custom voice id.

Provider Support

Provider	Models
OpenAI	gpt-4o-mini-tts (default), tts-1, tts-1-hd
OpenRouter	Speech models it lists, through its OpenAI-compatible endpoint (no default model)

Model Selection

const response = await ai.speech.get(
  { text: 'Hello world' },
  { metadata: { model: 'gpt-4o-mini-tts' } }
);

Chat models with the audio capability can also speak their answer; see modalities and audio on chat requests.