Skip to content

Kokoro-82M

An 82M-parameter text-to-speech model with voice packs

Visit official site

What Kokoro-82M is

The model is an open-source text-to-speech model created by hexgrad and hosted on Hugging Face, with 82 million parameters under the Apache 2.0 license. It released v0.19 on 25 December 2024 with 10 unique voice packs.

What you can do with it

  • Convert American and British English text into speech
  • Use voice packs to produce different speech styles
  • Generate natural speech output for a voice assistant
  • Turn text into spoken audio for audiobooks and news broadcasts

Getting started

  • Install needed libraries such as espeak-ng and phonemizer in Google Colab
  • Clone the model repository from Hugging Face
  • Build the model and load the default voice pack with the provided script
  • Call the generate function with text and a voice pack to produce 24khz audio

Model notes

  • The v0.19 version was released on 25 December 2024
  • 10 unique voice packs are provided
  • The material states it ranked first in the TTS Spaces Arena
  • ONNX format and an API with documentation support deployment

Pros & cons

✓ What we like

  • High-quality speech synthesis with few parameters and little data
  • An API and documentation are provided for developer integration

! What to watch out for

  • Named language support centres on American and British English
  • Running the documented steps requires a Google Colab setup

FAQ

What is the model size and license?

The material states 82 million parameters and the Apache 2.0 license.

What do the documented steps call?

The generate function takes text and a voice pack and produces 24khz audio with phonemes.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review