kokoro

What it is and what it does

Kokoro is a Python wrapper around the Kokoro-82M text-to-speech model, a lightweight neural network with 82 million parameters designed to generate natural-sounding speech from text. It handles phoneme conversion, voice synthesis, and audio generation in a single pipeline, supporting multiple languages (American English, British English, Spanish, French, Hindi, Italian, Japanese, Brazilian Portuguese, and Mandarin Chinese) through language-specific codes and optional misaki extensions.

The library is built for both interactive use (Jupyter notebooks, Google Colab) and production deployment. It depends on torch for inference, transformers for model loading, huggingface-hub for downloading weights, and misaki for grapheme-to-phoneme conversion. Audio output is generated at 24000 Hz sample rate. The model weights are Apache-licensed, making it suitable for commercial and personal projects without licensing friction.

Use it for:

• Generate speech from long-form text in Jupyter notebooks or Colab for prototyping and testing voice synthesis.
• Build a multilingual chatbot or voice assistant that speaks in multiple languages with preset voice profiles.
• Create audiobook or podcast narration pipelines by splitting text and synthesizing each section with consistent voice.
• Deploy a lightweight TTS service in resource-constrained environments where model size and inference speed matter.
• Experiment with different voice profiles and languages without managing model weights or phoneme rules manually.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Kokoro is an inference library for the Kokoro-82M text-to-speech model, enabling you to generate spoken audio from text in multiple languages with a lightweight, open-weight neural network.

Yes, with conditions. Install kokoro if you need a lightweight, open-weight TTS model for prototyping or production use and can tolerate the aging maintenance status (last release 496 days ago). The Apache license is permissive, install friction is low, and the package has solid community adoption. However, verify that the preset voices and supported languages meet your needs, and be aware that torch and transformers are heavy dependencies. If you require active development or frequent updates, check the repository first.

Install

kokoro on PyPI

pip install kokoro

uv add kokoro

poetry add kokoro

Installing kokoro

Before you install

Low install friction with a pure-Python wheel. The package depends on torch, transformers, huggingface-hub, numpy, loguru, and misaki—all standard ML dependencies. Maintenance status is aging (last release 496 days ago), though the repository remains active with recent commits and substantial community interest (8421 stars).

License in practice

Apache License 2.0 is permissive and allows commercial use, modification, and redistribution with minimal restrictions. You can deploy this package in production or personal projects without licensing concerns.

Quickstart

pip install kokoro>=0.9.4
from kokoro import KPipeline
pipeline = KPipeline(lang_code='a')
generator = pipeline('Hello world', voice='af_heart')
for gs, ps, audio in generator:
    print(gs, ps)  # graphemes, phonemes
    # audio is a numpy array at 24000 Hz sample rate

Requires espeak-ng system library (install via apt-get on Linux, .msi installer on Windows, or conda). Python >=3.10,<3.13 only. torch and transformers are heavy dependencies; initial model download from Hugging Face Hub occurs on first use.

Verify before relying

• Whether voice cloning or custom voice loading is supported beyond the preset voices mentioned in examples.
• Inference speed and memory footprint on CPU-only systems or edge devices.
• Quality comparison to larger commercial TTS models in production scenarios.
• Support status and roadmap given the aging maintenance signal (last release 496 days ago).

Package facts

Evidence: kokoro-0.9.4-py3-none-any.whl

Tags

More Linguistic packages