Text to Speech MP3 (Open Voices, No API Key) – Kokoro TTS

Turn any text into natural-sounding speech as MP3 or WAV with Kokoro-82M, one of the best open-weight text-to-speech models (Apache-2.0). No OpenAI, ElevenLabs or Google key needed, nothing to install: paste text, pick a voice, get a downloadable audio file per text.

• 41 voices in 7 languages: US English, UK English, Spanish, French, Hindi, Italian, Brazilian Portuguese
• Voice blending: mix voices with weights, e.g. am_michael:0.6,am_onyx:0.4, for your own signature voice
• Bulk: send hundreds of texts in one run, one audio file per text
• Long texts: up to 20,000 characters per text by default (configurable), split at sentence boundaries and joined seamlessly into one file
• Speed control from 0.7 (slow) to 1.3 (fast)
• MP3 (64–192 kbps) or WAV (16-bit PCM, 24 kHz)
• Public download link for every file, ready for your app, CMS, podcast feed or automation
• $0.02 per 1,000 characters, failed or empty texts are free

Who is it for?

• Content creators and marketers: audio versions of blog posts, newsletters and social videos
• E-learning: narrate lessons, flashcards and quizzes
• Developers and no-code builders: a cheap TTS API for apps, n8n/Make/Zapier flows and AI agents (MCP)
• Accessibility: listenable versions of articles and product pages
• Video makers: voice-overs for explainers, shorts and slides

Why this actor?

	This actor	Paid TTS API wrappers	Other Kokoro actors
API key needed	No	Yes (yours) or hidden markup	No
Price per 1,000 characters	$0.02	$0.04 – $0.40	$0.09 – $0.10
Start fee	None	often	sometimes
Voice blending	Yes	rarely	rarely
Bulk texts, one file each	Yes	varies	varies

Kokoro runs entirely inside the actor (the model is baked into the Docker image), so your text is never sent to a third-party TTS service.

Input

Field	Description
text	One text to read aloud.
texts	Bulk list: [{"id": "intro", "text": "..."}] (plain strings also work). The id becomes the file name.
voice	Voice name or blend (default af_heart). The first letter sets the language.
speed	0.7 – 1.3, default 1.0.
format	mp3 (default) or wav.
mp3Bitrate	64k, 96k, 128k (default) or 192k, mono.
maxCharsPerText	Safety cap per text (default 20,000). Longer texts are cut at the last full sentence and flagged truncated: true.

Example:

{
  "texts": [
    {"id": "intro", "text": "Welcome to our product tour. In the next two minutes you will see everything that matters."},
    {"id": "outro", "text": "Thanks for listening. See you next time!"}
  ],
  "voice": "am_michael:0.6,am_onyx:0.4",
  "speed": 1.0,
  "format": "mp3"
}

Output

One dataset row per text; the audio file itself is stored in the run's key-value store (<id>.mp3 / <id>.wav) and linked in audioUrl.

{
  "id": "text",
  "voice": "af_heart",
  "language": "English (US)",
  "speed": 1,
  "format": "mp3",
  "characters": 77,
  "durationSeconds": 5.03,
  "sizeBytes": 81836,
  "audioUrl": "https://api.apify.com/v2/key-value-stores/mCHDdV98wNH6uet4c/records/text.mp3?signature=qPTCmKxQS2jdkK2fv37l",
  "key": "text.mp3",
  "truncated": false,
  "status": "ok",
  "error": null,
  "textPreview": "Hello! This is Kokoro, an open text to speech model, reading your text aloud."
}

status is ok, error (e.g. empty text, not charged) or skipped (your max charge limit was reached).

Voices and languages

Only voices the engine really supports are accepted (Japanese and Chinese Kokoro voices are not supported in this actor).

Language	Female	Male
English (US) – a	af_heart, af_alloy, af_aoede, af_bella, af_jessica, af_kore, af_nicole, af_nova, af_river, af_sarah, af_sky	am_adam, am_echo, am_eric, am_fenrir, am_liam, am_michael, am_onyx, am_puck, am_santa
English (UK) – b	bf_alice, bf_emma, bf_isabella, bf_lily	bm_daniel, bm_fable, bm_george, bm_lewis
Spanish – e	ef_dora	em_alex, em_santa
French – f	ff_siwis	–
Hindi – h	hf_alpha, hf_beta	hm_omega, hm_psi
Italian – i	if_sara	im_nicola
Portuguese (BR) – p	pf_dora	pm_alex, pm_santa

Tips: af_heart, af_bella, am_michael, am_fenrir, bf_emma and bm_george are among the most natural. English pronunciation uses the misaki lexicon with an espeak-ng fallback for unknown words; the other languages use espeak-ng phonemes. You can force a pronunciation in English with inline phonemes: [Kokoro](/kˈOkəɹO/). Blend only voices of the same language; the first voice decides the language.

Pricing

Pay per event: $0.02 per started 1,000 characters of successfully synthesized text (event 1k-characters), counted per text. A 77-character sentence costs $0.02, a 2,400-character article $0.06, a 20,000-character chapter $0.40. Empty or failed texts are free, and there is no start fee. Default memory is 4 GB: in our test it synthesized 2.2 minutes of audio in 108 s (2 GB took 209 s at the same compute cost).

Speed: roughly 1.5–2x faster than real time, i.e. about 30–50 seconds of processing per 1,000 characters (about 70 seconds of audio).

API, n8n and MCP

Run it synchronously and get the rows back:

curl -X POST "https://api.apify.com/v2/acts/fast_data_asap~text-to-speech-kokoro/run-sync-get-dataset-items?token=YOUR_APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"text": "Hello from the Apify API!", "voice": "bf_emma", "format": "mp3"}'

Then download the file from audioUrl.

• n8n / Make / Zapier: use the Apify integration ("Run Actor and get dataset items") with actor fast_data_asap/text-to-speech-kokoro.
• AI agents (MCP): add https://mcp.apify.com?tools=fast_data_asap/text-to-speech-kokoro to Claude, Cursor or any MCP client to let your agent speak.

Notes

• Model: Kokoro-82M v1.0 (Apache-2.0) via ONNX Runtime on CPU, voice pack v1.0.
• Audio is 24 kHz mono. Files stay in the run's key-value store according to your Apify data retention.
• Please only synthesize text you have the rights to use, and do not use the voices to impersonate real people.