kenkui 2.0.0

Convert Ebooks to Audiobooks with [custom] voice samples

kenkui

Ebook-to-audiobook conversion engine. No GPU. No nonsense.

kenkui is a Python library that converts ebooks into high-quality M4B audiobooks using Kyutai's pocket-tts, running entirely on CPU.

Looking for the interactive CLI? Install kentui — it's the interactive front-end built on top of this library.

---

Install

pip install kenkui

Or with uv:

uv add kenkui

---

Quick Start

import kenkui

# Load config (creates default at ~/.config/kenkui/config.toml on first run)
config = kenkui.load_config()

# Build a ProcessingConfig
proc = kenkui.ProcessingConfig(
    ebook_path="book.epub",
    output_path=".",
    voice="alba",
)

# Run the conversion
ok = kenkui.run_job(proc)

---

Features

• Freaky fast M4B audiobook generation — 100% CPU, no GPU
• Multithreaded chapter processing
• Supports EPUB, MOBI/AZW/AZW3/AZW4, and FB2
• Multi-voice narration via NLP speaker attribution (Ollama, Anthropic, OpenAI, Google)
• Voice pool template for automatic voice assignment by role + gender + rank
• Chapter-voice mode: distinct voice per chapter
• Broadcast-quality audio post-processing chain
• Credits chapter: synthesized audio appended to every m4b
• Flexible chapter selection (presets + manual override)
• Series support: cross-book character roster with pinned voice assignments

---

API Reference

Config

config = kenkui.load_config()                  # default config
config = kenkui.load_config("fast-mode")       # named config
config = kenkui.load_config("/path/to/cfg.toml")

kenkui.save_config(config)
kenkui.save_config(config, "fast-mode")

names: list[str] = kenkui.list_configs()

Book parsing

result = kenkui.parse_book("book.epub")
# result.chapters, result.metadata, result.book_hash

filtered = kenkui.filter_chapters(result.book_hash, selection)

NLP

# Stage 1-2: entity scan + character clustering
scan = kenkui.fast_scan("book.epub", nlp_model="llama3.2")
# scan.characters: list[CharacterInfo]

# Stages 1-4: full pipeline with speaker attribution
result = kenkui.full_analysis(
    "book.epub",
    nlp_model="llama3.2",
    progress_callback=lambda pct, msg: print(f"{pct}% {msg}"),
)
# result.characters, result.chapters (annotated)

Voices

voices = kenkui.list_voices()
voices = kenkui.list_voices(gender="Female", accent="British", source="compiled")

voice = kenkui.get_voice("alba")   # VoiceInfo | None

cast = kenkui.suggest_cast(
    roster=scan.characters,
    excluded_voices=["alba"],
    default_voice="sarah",
)
# cast.speaker_voices: dict[str, str]

narrator = kenkui.recommend_narrator(scan.characters, default_voice="alba")

result = kenkui.exclude_voice("marius")   # ExcludeResult
result = kenkui.include_voice("marius")   # IncludeResult

preview = kenkui.audition_voice("alba", text="Hello world.")
# preview.audio_path, preview.duration_ms

dl = kenkui.download_voice(force=False)   # DownloadResult
dl = kenkui.fetch_voice(repo_id="user/repo")

Series

series_list = kenkui.list_series()
entry = kenkui.get_series("my-series")
entry = kenkui.create_series("My Series")
kenkui.update_series(entry)

HuggingFace auth

result = kenkui.authenticate_huggingface("hf_token_here")
# result.authenticated, result.username

Job runner

from kenkui import ProcessingConfig, NarrationMode

config = ProcessingConfig(
    ebook_path="book.epub",
    output_path=".",
    voice="alba",
    narration_mode=NarrationMode.SINGLE,
    chapter_filters=[FilterOperation(type="preset", value="content-only")],
)

ok: bool = kenkui.run_job(config)
ok: bool = kenkui.run_job(
    config,
    progress_callback=lambda pct, chapter, eta: print(f"{pct:.0f}% {chapter}"),
)

---

Voice System

Voices come in three tiers:

Tier	Source	Auth required?
Compiled	Downloaded from HuggingFace on first run	No
Built-in	8 pocket-tts defaults	No
Custom	.wav files (user-provided or fetched)	Yes (HuggingFace)

Built-in voices: alba, marius, javert, jean, fantine, cosette, eponine, azelma

---

Configuration

kenkui uses TOML config files stored at ~/.config/kenkui/ (XDG).

Key	Default	Description
workers	cpu_count - 2	Parallel TTS worker processes
m4b_bitrate	96k	Output audio bitrate
temp	0.7	Sampling temperature
lsd_decode_steps	1	LSD decode steps
default_voice	alba	Fallback voice
nlp_provider	ollama	NLP backend
nlp_model	llama3.2	Model for speaker attribution
credits_enabled	true	Append synthesized credits audio

---

Non-Goals

kenkui is not a general-purpose TTS framework, a GUI app, or an MP3 generator. The focus is narrow: fast, high-quality audiobook generation from ebooks.

---

🙏 Special Thanks

Thanks to Project Gutenberg for providing some of the public-domain books included with kenkui.

---

Voice Dataset Credits

kenkui's compiled voices are derived from two publicly available speech corpora.

CSTR VCTK Corpus

Veaux, Christoph; Yamagishi, Junichi; MacDonald, Kirsten. (2019). CSTR VCTK Corpus: English Multi-speaker Corpus for CSTR Voice Cloning Toolkit. University of Edinburgh. The Centre for Speech Technology Research (CSTR).

Licensed under Creative Commons Attribution 4.0 (CC BY 4.0). Commercial use is permitted with attribution.

EARS Dataset

Licensed under Creative Commons Attribution-NonCommercial 4.0 (CC BY-NC 4.0).

Note: Compiled voices sourced from EARS (identifiable by EARS in the voice name via kenkui.list_voices()) may not be used for commercial purposes. If you are building a commercial product with kenkui, use only VCTK-sourced or built-in voices.