Kokoro TTS

State-of-the-art AI Text-to-Speech Model

Try UI

Exploring Kokoro TTS: A Powerful Local Text-to-Speech Solution

As the demand for voice applications continues to grow, many developers are seeking robust local text-to-speech (TTS) systems that eliminate the need to rely on external APIs like OpenAI, Google, or ElevenLabs. One standout option is Kokoro TTS, a lightweight and high-performing TTS model that has gained significant attention for its capabilities and accessibility.

What is Kokoro TTS?

Kokoro TTS is a compact yet powerful text-to-speech model, currently available on Hugging Face and GitHub. Despite its modest size—trained on less than 100 hours of audio—it delivers impressive results, consistently topping the TTS leaderboard on Hugging Face. Unlike larger systems, Kokoro TTS offers the advantage of running locally, even on devices without GPUs, making it accessible for a wide range of users.

Key Features

1. Multi-Language and Voice Support

Kokoro TTS includes a variety of voices across different languages, including American and British English, French, Japanese, Korean, and Chinese. Users can explore these voices and even create new ones by blending or customizing existing voice embeddings.

2. Custom Voice Creation

Each voice in Kokoro TTS is associated with a unique embedding. By blending these embeddings, users can create new, personalized voices. Techniques such as weighted averaging or spherical interpolation allow for precise control over the resulting voice characteristics.

3. Open Source and Community-Driven

Kokoro TTS has inspired the creation of numerous related projects, such as:

  • Kokoro Onnx: A package optimized for fast, local inference using Onnx models.
  • Kokoro FastAPI TTS: A tool that emulates OpenAI-compatible speech endpoints, making it easy to integrate Kokoro TTS into existing applications.

4. Ease of Use

The system is straightforward to set up, with detailed examples and support for popular tools like Colab and virtual environments. This accessibility lowers the barrier for developers looking to integrate TTS capabilities into their projects.

Real-World Applications

Kokoro TTS is ideal for developers and enthusiasts aiming to build local voice-enabled applications without incurring API costs. It pairs seamlessly with automatic speech recognition (ASR) systems to create local conversational agents, making it suitable for privacy-focused or offline applications.

Getting Started

Setting up Kokoro TTS involves downloading the model and embeddings, running the system locally with tools like Kokoro Onnx, and customizing voices as needed. Whether you're generating audio for a project or experimenting with voice synthesis, Kokoro TTS offers a flexible and cost-effective solution.

Why Choose Kokoro TTS?

Kokoro TTS stands out not just for its quality but also for its simplicity and flexibility. It's an excellent choice for those seeking a lightweight, local TTS solution without compromising on performance or scalability.

Explore Kokoro TTS today to unlock new possibilities in text-to-speech technology!

Key Features

82M Parameters

Efficient model with only 82 million parameters, outperforming larger models.

Multiple Voicepacks

10 unique voicepacks available, with more to come.

#1 Ranked Model

Topped the TTS Spaces Arena, outperforming models with more parameters and data.

Quick Start

OpenAI-Compatible Speech Endpoint

Using OpenAI's Python library

from openai import OpenAI
client = OpenAI(base_url="https://api.kokorotts.com/v1", api_key="not-needed")
response = client.audio.speech.create(
    model="kokoro",  # Not used but required for compatibility, also accepts library defaults
    voice="af_bella+af_sky",
    input="Hello world!",
    response_format="mp3"
)

response.stream_to_file("output.mp3")

Using Requests

import requests

response = requests.post(
    "https://api.kokorotts.com/v1/audio/speech",
    json={
        "model": "kokoro",  # Not used but required for compatibility
        "input": "Hello world!",
        "voice": "af_bella",
        "response_format": "mp3",  # Supported: mp3, wav, opus, flac
        "speed": 1.0
    }
)

# Save audio
with open("output.mp3", "wb") as f:
    f.write(response.content)