Generate speech audio files from text using HeyGen's in-house Starfish TTS model via the v3 API. This skill is for standalone audio generation — separate from video creation.

Authentication

All requests require the X-Api-Key header. Set the HEYGEN_API_KEY environment variable.

curl -X GET "https://api.heygen.com/v3/voices?engine=starfish" \
  -H "X-Api-Key: $HEYGEN_API_KEY"

Tool Selection

If HeyGen MCP tools are available (mcp__heygen__*), prefer them over direct HTTP API calls.

Default Workflow

• List voices with mcp__heygen__list_audio_voices (or GET /v3/voices?engine=starfish)
• Pick a voice matching desired language, gender, and features
• Call mcp__heygen__text_to_speech (or POST /v3/voices/speech) with text and voice_id
• Use the returned audio_url to download or play the audio

List TTS Voices

Retrieve voices compatible with the Starfish TTS model.

Note: This uses the unified GET /v3/voices endpoint with the engine=starfish filter to return only TTS-compatible voices. Not all video voices support Starfish TTS. The response is paginated — use next_token to fetch additional pages.

Query Parameters

curl

curl -X GET "https://api.heygen.com/v3/voices?engine=starfish" \
  -H "X-Api-Key: $HEYGEN_API_KEY"

TypeScript

interface AudioVoiceItem {
  voice_id: string;
  name: string;
  language: string;
  gender: "female" | "male" | "unknown";
  preview_audio_url: string | null;
  support_pause: boolean;
  support_locale: boolean;
  type: string;
}
interface TTSVoicesResponse {
  error: null | string;
  data: AudioVoiceItem[];
  has_more: boolean;
  next_token: string | null;
}
async function listTTSVoices(): Promise {
  const allVoices: AudioVoiceItem[] = [];
  let token: string | null = null;
  do {
    const url = new URL("https://api.heygen.com/v3/voices");
    url.searchParams.set("engine", "starfish");
    if (token) url.searchParams.set("token", token);
    const response = await fetch(url.toString(), {
      headers: { "X-Api-Key": process.env.HEYGEN_API_KEY! },
    });
    const json: TTSVoicesResponse = await response.json();
    if (json.error) {
      throw new Error(json.error);
    }
    allVoices.push(...json.data);
    token = json.next_token;
  } while (token);
  return allVoices;
}

Python

import requests
import os
def list_tts_voices() -> list:
    all_voices = []
    token = None
    while True:
        params = {"engine": "starfish"}
        if token:
            params["token"] = token
        response = requests.get(
            "https://api.heygen.com/v3/voices",
            headers={"X-Api-Key": os.environ["HEYGEN_API_KEY"]},
            params=params,
        )
        data = response.json()
        if data.get("error"):
            raise Exception(data["error"])
        all_voices.extend(data["data"])
        if not data.get("has_more"):
            break
        token = data.get("next_token")
    return all_voices

Response Format

{
  "error": null,
  "data": [
    {
      "voice_id": "f38a635bee7a4d1f9b0a654a31d050d2",
      "name": "Chill Brian",
      "language": "English",
      "gender": "male",
      "preview_audio_url": "https://resource.heygen.ai/text_to_speech/WpSDQvmLGXEqXZVZQiVeg6.mp3",
      "support_pause": true,
      "support_locale": false,
      "type": "public"
    }
  ],
  "has_more": false,
  "next_token": null
}

Generate Speech Audio

Convert text to speech audio using a specified voice.

Endpoint

POST https://api.heygen.com/v3/voices/speech

Request Fields

curl

curl -X POST "https://api.heygen.com/v3/voices/speech" \
  -H "X-Api-Key: $HEYGEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Hello! Welcome to our product demo.",
    "voice_id": "YOUR_VOICE_ID",
    "speed": 1.0
  }'

TypeScript

interface TTSRequest {
  text: string;
  voice_id: string;
  input_type?: "text" | "ssml";
  speed?: number;
  language?: string;
  locale?: string;
}
interface WordTimestamp {
  word: string;
  start: number;
  end: number;
}
interface TTSResponse {
  error: null | string;
  data: {
    audio_url: string;
    duration: number;
    request_id?: string;
    word_timestamps?: WordTimestamp[];
  };
}
async function textToSpeech(request: TTSRequest): Promise {
  const response = await fetch(
    "https://api.heygen.com/v3/voices/speech",
    {
      method: "POST",
      headers: {
        "X-Api-Key": process.env.HEYGEN_API_KEY!,
        "Content-Type": "application/json",
      },
      body: JSON.stringify(request),
    }
  );
  const json: TTSResponse = await response.json();
  if (json.error) {
    throw new Error(json.error);
  }
  return json.data;
}

Python

import requests
import os
def text_to_speech(
    text: str,
    voice_id: str,
    input_type: str = "text",
    speed: float = 1.0,
    language: str | None = None,
    locale: str | None = None,
) -> dict:
    payload = {
        "text": text,
        "voice_id": voice_id,
        "speed": speed,
    }
    if input_type != "text":
        payload["input_type"] = input_type
    if language:
        payload["language"] = language
    if locale:
        payload["locale"] = locale
    response = requests.post(
        "https://api.heygen.com/v3/voices/speech",
        headers={
            "X-Api-Key": os.environ["HEYGEN_API_KEY"],
            "Content-Type": "application/json",
        },
        json=payload,
    )
    data = response.json()
    if data.get("error"):
        raise Exception(data["error"])
    return data["data"]

Response Format

{
  "error": null,
  "data": {
    "audio_url": "https://resource2.heygen.ai/text_to_speech/.../id=365d46bb.wav",
    "duration": 5.526,
    "request_id": "p38QJ52hfgNlsYKZZmd9",
    "word_timestamps": [
      { "word": "", "start": 0.0, "end": 0.0 },
      { "word": "Hey", "start": 0.079, "end": 0.219 },
      { "word": "there,", "start": 0.239, "end": 0.459 },
      { "word": "", "start": 5.526, "end": 5.526 }
    ]
  }
}

Usage Examples

Basic TTS

const result = await textToSpeech({
  text: "Welcome to our quarterly earnings call.",
  voice_id: "YOUR_VOICE_ID",
});
console.log(Audio URL: ${result.audio_url});
console.log(Duration: ${result.duration}s);

With Speed Adjustment

const result = await textToSpeech({
  text: "We're thrilled to announce our newest feature!",
  voice_id: "YOUR_VOICE_ID",
  speed: 1.1,
});

With Language and Locale for Multilingual Voices

const result = await textToSpeech({
  text: "Bem-vindo ao nosso produto.",
  voice_id: "MULTILINGUAL_VOICE_ID",
  language: "pt",
  locale: "pt-BR",
});

With SSML Input

const result = await textToSpeech({
  text: 'Hello  and welcome!',
  voice_id: "YOUR_VOICE_ID",
  input_type: "ssml",
});

Find a Voice and Generate Audio

async function generateSpeech(text: string, language: string): Promise {
  const voices = await listTTSVoices();
  const voice = voices.find(
    (v) => v.language.toLowerCase().includes(language.toLowerCase())
  );
  if (!voice) {
    throw new Error(No TTS voice found for language: ${language});
  }
  const result = await textToSpeech({
    text,
    voice_id: voice.voice_id,
  });
  return result.audio_url;
}
const audioUrl = await generateSpeech("Hello and welcome!", "english");

Pauses with Break Tags

Use SSML-style break tags in your text for pauses:

word  word

Rules:

• Use seconds with s suffix:
• Must have spaces before and after the tag
• Self-closing tag format

With v3, you can also use input_type: "ssml" for full SSML support, allowing richer markup beyond just break tags:

{
  "text": "Welcome!  Let's get started.",
  "voice_id": "YOUR_VOICE_ID",
  "input_type": "ssml"
}

Best Practices

• Use GET /v3/voices?engine=starfish to find compatible voices — the unified /v3/voices endpoint serves all voice types, so the engine=starfish filter is essential for TTS
• Check support_locale before setting a locale — only multilingual voices support locale selection
• Keep speed between 0.8-1.2 for natural-sounding output
• Preview voices using the preview_audio_url before generating (may be null for some voices)
• Use word_timestamps in the response for caption syncing or timed text overlays
• Use SSML break tags in your text for pauses: word word
• Use input_type: "ssml" when you need full SSML markup control beyond simple break tags
• Paginate voice listing — the v3 endpoint returns paginated results; use has_more and next_token to fetch all voices