AI GitHub
Kokoro for compact text-to-speech pipelines
Generate speech locally using published Kokoro voices and inference tooling.
Key Takeaways
- Kokoro combines text preprocessing, language-dependent phonemization and a synthesis model.
- Choose a voice, language configuration and compatible phonemizer.
- Read a short passage containing a date, an abbreviation and a name.
- Supported language and voice combinations depend on the chosen assets.
How it works
Kokoro combines text preprocessing, language-dependent phonemization and a synthesis model. Voice selection and text normalization affect the sound of the same input.
What you need
Choose a voice, language configuration and compatible phonemizer. If text arrives dynamically, preserve the exact normalized sentence used to generate any audio you review.
What to try first
Read a short passage containing a date, an abbreviation and a name. Inspect pronunciation and split longer input into manageable segments.
Compare a natural paragraph break with automatic segmentation at a shorter limit. Pacing and intelligibility can change at the boundaries between synthesized parts.
Where to look for failures
Supported language and voice combinations depend on the chosen assets. Check phonemizer dependencies and verify terms for each model and voice.
Related Franklin pages
Compare on-device synthesis with several short phrases that contain names and numbers. Log initialization time separately from steady-state speech generation.
- Run neural text-to-speech locally with downloadable voice models. Piper for local speech synthesis on small devices
- Embed supported speech models in applications through ONNX-based runtimes. sherpa-onnx for local speech applications
For an application where the quality of recording and handling audio matters, see Franklin’s Open Mic AI profile. Open Mic AI
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗