← Back to all models

Kokoro-TTS

hexgradAudio & SpeechOpen Source

Lightweight open-source TTS model with 82M parameters delivering quality comparable to commercial models. Supports 9 languages, 54+ voices, and runs on both GPU and CPU.

Abilities

Audio GenerationMultilingualOn-device

Use Cases

CreativeAutomationOn-device

Available in Tools

Kokoro-FastAPIhttps://github.com/remsky/Kokoro-FastAPI

Availability

Hugging Facehttps://huggingface.co/hexgrad/Kokoro-82M
Self-hosted
pip installhttps://github.com/hexgrad/kokoro

How to Use

Install via `pip install kokoro soundfile` (requires espeak-ng). Run locally on CPU or GPU. OpenAI-compatible API available through community wrappers like Kokoro-FastAPI.

Pros

  • Fully open source (Apache 2.0) — free for commercial use
  • Only 82M parameters — runs fast on CPU, no GPU required
  • Quality comparable to much larger commercial models
  • 54+ voices with voice blending and speed adjustment
  • 9 languages including English, Spanish, French, Japanese, Chinese
  • Fully offline — no API calls, no data leaves your machine

Cons

  • Only 9 languages — fewer than commercial alternatives with 30+
  • 24kHz sample rate — lower than some commercial offerings
  • No voice cloning — limited to included voices and blending
  • Requires espeak-ng system dependency
  • Less emotionally expressive than Hume or ElevenLabs

What to Use It For

star

Perfect For

Self-hosted TTS applications

Apache 2.0 license, 82M params runs on CPU — perfect for on-premise TTS without API costs

Privacy-sensitive TTS

Fully offline — no data sent to external servers, runs entirely on your hardware

thumb_up

Good For

Prototyping voice applications

Free, fast, and easy to set up — great for testing before committing to paid services

warning

Not Recommended

Production with 30+ languages

Only 9 languages supported — not enough for global products

Try instead: ElevenLabs

Emotionally nuanced speech

Basic expressiveness — lacks fine-grained emotion control

Try instead: Hume

block

Do Not Use For

Speech-to-text

TTS only — generates speech, does not transcribe

Try instead: Whisper

Voice cloning

No voice cloning capability — fixed voice set

Try instead: ElevenLabs

Technical Details

PricingFree (open source)
Parameters82M
LicenseApache 2.0

Links

language Website description API Documentation code Repository