Today, I fired up the Voices macOS app, which I occasionally use to convert blog posts or documentation text to audio files that I can take on the go. I usually use one of OpenAI’s Text to speech APIs, but today I noticed there was a new option called Kokoro, which is a local voice model. I was interested in the model and tried out a few voices. Two or three of them were amazing for a local model.

After trying out a few blog posts and listening to them, I decided to find a Python version of the model. It might be slow to run, but at least I can create a pipeline to automate the process of generating voice files from arbitrary text files I have lying around.

That’s when I found the Kokoro TTS application. I downloaded it and noticed that it needed a few things. Things like NumPy required to be installed, and a couple of things were out of sync, but the project mostly just worked. While it wasn’t quite as fast as a native Mac app, it was still good enough for me to run text files through it very quickly.

Tonight, I decided to write up my notes on how to bootstrap and run the project.

Installation

git clone git@github.com:nazdridoy/kokoro-tts.git

cd kokoro-tts

uv add numpy

uv sync 

wget https://github.com/nazdridoy/kokoro-tts/releases/download/v1.0.0/voices-v1.0.bin

wget https://github.com/nazdridoy/kokoro-tts/releases/download/v1.0.0/kokoro-v1.0.onnx

Basic Usage

Once installed, you can use Kokoro TTS from the command line:

echo "Hello, this is a test of Kokoro TTS" | ./kokoro-tts /dev/stdin example-streaming.mp3 \
    --format mp3 \
    --voice af_heart

If you already have an existing text file:

./kokoro-tts textfile.txt example-textfile.mp3 \
    --format mp3 \
    --voice af_heart

Conclusion

Kokoro TTS is a solid local text-to-speech option that runs entirely on your machine without requiring API calls or internet connectivity. While it may not be as fast as native Mac apps, the quality is impressive for a local model, and the ability to automate voice file creation from text files makes it valuable for anyone who wants to convert written content to audio format.

One major advantage I discovered is that unlike most cloud APIs like OpenAI’s Text-to-speech that limit posts to around 3000 characters, Kokoro didn’t seem to have an upper limit on any of the content I tried it with. This makes it particularly useful for longer blog posts and documentation.

The setup process is straightforward once you handle the missing dependencies, and the command-line interface makes it easy to integrate into existing workflows or automation scripts.

Written by Jeff. Edited with Grammarly and Claude Code.