Skip to content
Search

Kokoro

lastmod 2025-01-17created 2025-01-171 edits

Kokoro-82M

Converting e-books into audiobooks with the model

  • Converting e-books into audiobooks with the Kokoro-82M model | GeekNews
  • Kokoro v0.19 is a recently released text-to-speech model with 82M parameters that delivers very high-quality output
    • Apache license, trained on less than 100 hours of audio
    • Supports American English, British English, French, Korean, Japanese, and Chinese, and offers a variety of high-quality voices
  • Using Kokoro
    • Users can use Kokoro through a tool called Audiblez, which converts e-books into audiobooks.
    • Audiblez parses .epub files and converts the body text of the book into well-recorded audio files.
    • For example, converting a book of about 100,000 words takes about 2 hours on an M2 MacBook Pro.
  • Installation and running
    • Audiblez can be installed with pip on a computer with Python 3 installed.
    • Does not work on Python 3.13.
    • About 360MB of additional files must be downloaded.
    • To convert an .epub file into an audiobook, run the command.
  • Supported languages and voices
    • The language can be specified with the -l option; supported language codes are en-us, en-gb, fr-fr, ja, ko, cmn.
    • The voice can be specified with the -v option, and a variety of voices are offered.
  • Chapter detection
    • Chapter detection is a bit unreliable, but it can find the core chapters in most .epub files.
    • If a chapter of interest is not included, you can try adjusting the is_chapter function in the code.
  • Source code and improvements
    • The Audiblez project is available on GitHub.
    • Future improvements include better chapter detection, adding chapter navigation, and adding narration for images.

See also

author Zer0 sitemap