AI Text to Speech – 49 Natural Voices, 10 Languages

Turn any text into natural-sounding speech in seconds. Pick one of 49 ready-made voices across 10 languages, optionally describe the emotion or tone you want in plain words, and get back a downloadable WAV or MP3 file — all in a single API call.

No signup, no API key management, no audio tools to install. You only pay for Apify compute.

What you get

• 49 natural voices — including regional dialects (Cantonese, Sichuan, Beijing, Shanghai) plus native-quality English, Spanish, Russian, Italian, Korean, Japanese, German, French and Portuguese voices.
• 10 languages with automatic detection — or pin a language for the best pronunciation.
• Emotion & tone control — write a short instruction like "Speak slowly and calmly, like a bedtime storyteller" and the voice follows it.
• Long text in one call — up to 5,000 characters as a single, seamless clip.
• WAV or MP3 output — ready to download from a direct link, ideal for voiceovers, audiobooks, apps and chatbots.
• Built-in reliability — if one route is busy, the Actor automatically retries another, so runs rarely fail.

Quick start

{
  "text": "Hello! Welcome to our channel.",
  "voice": "Vivian",
  "language": "Auto",
  "format": "mp3"
}

Press Start (or call the API), and the run's dataset gives you a direct download link:

{
  "audio_url": "https://api.apify.com/v2/key-value-stores/<store-id>/records/audio.mp3",
  "voice": "Vivian",
  "chars": 31,
  "duration_s": 3.2,
  "format": "mp3"
}

Control the emotion and style

Add the optional instruct field to steer emotion, speed and character with plain language:

{
  "text": "I just heard the funniest joke. You would not believe it.",
  "voice": "Ryan",
  "language": "English",
  "instruct": "Very happy, upbeat and energetic."
}

Good instructions are short and descriptive:

• Whisper softly, as if telling a secret in a quiet room.
• Speak angrily, with a sharp and forceful tone.
• Slow and calm, like a bedtime storyteller.
• 用特别愤怒的语气说

Notes when using instruct:

• Instructions work best in English and Chinese; other languages may be less stable.
• Style generation uses a high-demand route, so it can occasionally be unavailable. If that happens, the run fails with a clear message instead of silently ignoring your instruction — just try again in a few minutes.
• Turn on Rotate IP via Apify Proxy for high-volume jobs.

Input fields

Field	Type	Default	Description
text	string	required	Text to synthesize, up to 5,000 characters.
voice	enum	Vivian	One of 49 voices (see the dropdown).
language	enum	Auto	Auto, English, Chinese, German, Italian, Portuguese, Spanish, Japanese, Korean, French, Russian.
format	enum	wav	wav (24 kHz PCM) or mp3 (128 kbps).
instruct	string	optional	Plain-language style/emotion instruction. Leave empty for the standard high-stability mode.
useProxy	boolean	false	Route requests through Apify Proxy for high-volume jobs (billed separately by Apify).
browserFallback	boolean	true	Keep on for maximum reliability — enables an automatic backup route.
browserTimeoutSecs	integer	180	Maximum time for a single backup attempt.
requireInstruct	boolean	true	When instruct is set, fail clearly if the style route is unavailable instead of returning plain speech.

Run with the Apify API

curl -X POST \
  "https://api.apify.com/v2/acts/andrew_babo~qwen3-tts-api/runs?token=$APIFY_TOKEN&waitForFinish=300" \
  -H "Content-Type: application/json" \
  -d '{"text":"Hello from Qwen3 TTS.","voice":"Ryan","language":"English","format":"mp3"}'

Tips

• Keep language on Auto unless pronunciation sounds off — explicit selection usually fixes it.
• One long call is faster and seam-free compared to splitting text into many small requests.
• Typical run time is just a few seconds for standard voices, and 15–45 seconds when using style instructions.

Use cases

• YouTube, TikTok and podcast voiceovers
• Audiobooks and e-learning narration
• In-app voice, chatbots and IVR prompts
• Multilingual content localization

Pricing

The Actor itself is free — you only pay Apify platform compute (a 4 GB run costs a fraction of a cent per generation).

License

The Actor code is Apache-2.0. The underlying Qwen3-TTS model is governed by its own license — review it before commercial use.