chatterbook 0.4.2

Convert serialized EPUB books into M4B audiobooks with Chatterbox Multilingual TTS.

chatterbook

Convert EPUB books into M4B audiobooks with Chatterbox Multilingual TTS.

from chatterbook import Book

book = Book("book.epub")
book.convert(
    language="ko",
    voice_path="voices/narrator.wav",
    style="warm",
    batch_size=8,
    speed=0.9,
)

Book reads the EPUB immediately and builds a JSON-compatible representation with title, chapters, paragraphs, narration/dialogue segments, and pause timing. You can serialize and restore that representation without reading the EPUB again:

data = book.to_dict()
book = Book.from_dict(data)
book.convert("audiobooks/book.m4b", language="ko")

By default, the output filename is read from the EPUB title metadata and written as Title.m4b in the current directory. You can also pass a path or directory.

voice_path is an optional short WAV reference clip for voice cloning. If it is omitted, Chatterbox's bundled default conditionals are used.

Chapters are split into paragraph and speech segments when the Book is built. Adjacent narration or dialogue sentences are compacted up to max_chars so TTS uses fewer calls while still keeping narration/dialogue boundaries. Quoted dialogue is marked separately and generated with slightly more expressive settings. Commas, sentence endings, paragraph breaks, and narration/dialogue transitions add short pauses to improve audiobook pacing.

The default audiobook pacing is intentionally slower than raw TTS:

• speed=0.9
• comma_pause_ms=120
• sentence_pause_ms=300
• paragraph_pause_ms=600
• dialogue_pause_ms=300

M4B output requires ffmpeg on your PATH. During conversion, chatterbook shows one colored tqdm progress bar for the whole EPUB. Pass show_progress=False to disable it.

When voice_path is used, conversion prepares voice conditionals once per same-style batch instead of once per segment. Tune batch_size to control how many adjacent narration or dialogue segments are grouped for that preparation step.

To export chapter WAV files instead of one M4B:

book.convert(
    "audio",
    language="ko",
    output_format="wav",
)

Styles

• neutral: balanced default
• warm: slightly softer narration
• dramatic: more expressive narration

You can override a style with explicit generation values:

book.convert(
    language="ko",
    exaggeration=0.7,
    cfg_weight=0.3,
)