Skip to content

kokoro.md

Latest commit

 

History

History
52 lines (38 loc) · 1.48 KB

kokoro.md

File metadata and controls

52 lines (38 loc) · 1.48 KB
Raw

Kokoro Engine

KokoroEngine wraps the Kokoro pipeline. It is a local neural engine with language-code-aware voice selection, voice formulas, and timing support.

Install

pip install "realtimetts[kokoro]"

For extra language stacks, install the matching RealtimeTTS extras:

pip install "realtimetts[kokoro,jp,zh,ko]"

Some non-English paths may also require OS-level espeak-ng support depending on the Kokoro package and platform.

Minimal Use

from RealtimeTTS import TextToAudioStream, KokoroEngine


if __name__ == "__main__":
    engine = KokoroEngine(voice="af_heart")
    stream = TextToAudioStream(engine)
    stream.feed("Hello from Kokoro.")
    stream.play()

Source Notes

  • Default voice is af_heart.
  • Voice prefixes determine language codes: af/am American English, bf/bm British English, jf/jm Japanese, zf/zm Mandarin Chinese, plus Spanish, French, Hindi, Italian, and Brazilian Portuguese prefixes.
  • Weighted voice formulas such as 0.3*af_sarah + 0.7*am_adam are parsed and cached.
  • Important controls include default_speed, silence trimming fields, and debug.
  • Zaphod dev-log benchmarks found Kokoro among the fastest balanced local baselines in that environment.

Troubleshooting

  • If a voice loads with the wrong language pipeline, pass a KokoroVoice with an explicit language_code.
  • Missing language package errors usually mean the corresponding jp, zh, or ko extras were not installed.