bookpeek

bookpeek scans the opening seconds of an audiobook and extracts the spoken title, title candidates, authors, and narrators without requiring an online AI service or API key. By default it runs Whisper and Vosk together, merges their extraction evidence, and keeps both raw transcripts under transcripts. Long pauses in the intro become hard sentence breaks so cues like read by … do not swallow the next section. Title candidates can come from Audible-style introductions, ID3 tags, filenames, folders, and chapter text, with scores indicating which candidate was preferred.

Online results are grouped by provider under online_matches, with separate authors and works arrays. Goodreads author results with zero similarity are omitted when any positive-scoring author exists; otherwise authors with at least ten works are retained as a fallback.

The result's top-level title, author, and narrators fields contain the final values after provider validation (narrators are de-duplicated case-insensitively with proper name casing). The nested extraction object remains the unmodified transcript/metadata extraction.

Audnexus results appear under online_matches.audnexus.works and include matched Audible identifiers, authors, narrators, region, and confidence scores. Verified narrator names are also collected under online_matches.audnexus.narrators.

Install

First-time setup

There are only two steps:

1. Install ffmpeg, which decodes audiobook files:

 # macOS with Homebrew
 brew install ffmpeg

 # Debian/Ubuntu
 sudo apt update && sudo apt install ffmpeg

2. Run the installer:

 ./install.sh

The installer bootstraps Poetry if necessary, installs Whisper and Vosk plus online enrichment, downloads spaCy's English model, and creates ~/.local/bin/bookpeek. ASR model weights are not downloaded during install; the first bookpeek scan prompts for them (or use -d to download automatically). If ffmpeg is missing, the installer prints the appropriate commands and exits. Add ~/.local/bin to your PATH if the bookpeek command is not found.

# Skip Open Library / Goodreads packages (ASR engines are still installed):
./install.sh --offline

On Apple Silicon, Whisper uses Metal automatically when device = "auto". Vosk is installed from the matching bookpeek GitHub Release wheel because upstream PyPI does not publish macOS arm64 wheels for 0.3.45. Whisper and Vosk model files download into ~/.config/bookpeek/models/ (and the usual Hugging Face cache for Whisper) on first use.

Usage

By default, bookpeek uses both Whisper and Vosk. Pass --engine whisper or --engine vosk to use only one backend. It scans 30 seconds first; if that only produces an Audible bumper such as This is Audible, it automatically retries at 60 seconds and then 90 seconds. A normal introduction stops after the first successful scan.

# Scan an audiobook file (Whisper + Vosk by default, then merge evidence):
bookpeek scan /path/to/book.m4b

# Scan a folder; bookpeek uses the first audio file alphabetically:
bookpeek scan /path/to/audiobook-folder/

# Use only the small CPU-friendly Vosk engine:
bookpeek scan /path/to/book.m4b --seconds 42 --engine vosk

# Force Whisper's tiny English model on the CPU:
bookpeek scan /path/to/book.m4b --engine whisper --model tiny.en --device cpu

# Choose Whisper and Vosk models independently for the default dual scan:
bookpeek scan /path/to/book.m4b --whisper-model base.en --vosk-model vosk-model-small-en-us-0.15

# Automatically download any missing ASR models (Vosk zip / Whisper weights).
# Vosk models are stored under ~/.config/bookpeek/models/ next to config.toml:
bookpeek scan /path/to/book.m4b -d

# Search (only) the specific Audible regions, in this order:
bookpeek scan /path/to/book.m4b --regions uk,ca,au

# Enable Open Library, Goodreads, and Audnexus enrichment for this scan:
bookpeek scan /path/to/book.m4b -w

# Force a completely offline scan, even if config enables enrichment:
bookpeek scan /path/to/book.m4b --offline

# Print the active configuration:
bookpeek config show

# Create a new config file with bookpeek's defaults:
# (Saves to ~/.config/bookpeek/config.toml; use --force to overwrite.)
bookpeek config new

Example response

The JSON result includes the extracted metadata and online provider matches:

{
  "path": "/path/to/The Blighted Stars.m4b",
  "title": "The Blighted Stars (The Devoured Worlds, #1)",
  "author": "Megan E. O'Keefe",
  "narrators": [
    "Ciaran Saward"
  ],
  "extraction": {
    "title": "The Blighted Stars",
    "title_candidates": [
      {
        "text": "The Blighted Stars",
        "source": "id3_title",
        "score": 0.9
      },
      {
        "text": "The Devoured Worlds 01 - The Blighted Stars",
        "source": "folder",
        "score": 0.65
      }
    ],
    "author": "Megan E. O'Keefe",
    "narrators": [
      "Kieran Sord"
    ]
  },
  "online_matches": {
    "audnexus": {
      "authors": [
        {
          "asin": "B00IMSE6VQ",
          "name": "Megan E. O'Keefe",
          "region": "us",
          "score": 1.0
        }
      ],
      "works": [
        {
          "asin": "B0BJ4JLW57",
          "title": "The Blighted Stars",
          "authors": ["Megan E. O'Keefe"],
          "narrators": ["Ciaran Saward"],
          "region": "us",
          "isbn": "9781668615331",
          "score": 0.933,
          "narrator_score": 0.667,
          "match_reason": "title_author"
        }
      ],
      "narrators": ["Ciaran Saward"]
    },
    "goodreads": {
      "authors": [],
      "works": [
        {
          "title": "The Blighted Stars (The Devoured Worlds, #1)",
          "title_complete": "The Blighted Stars (The Devoured Worlds, #1)",
          "author": "Megan E. O'Keefe",
          "score": 0.857
        }
      ]
    },
    "openlibrary": {
      "authors": [],
      "works": [
        {
          "title": "Devoured Worlds Series, 3-book collection ...",
          "author": "Megan E. O'Keefe",
          "score": 0.632
        }
      ]
    }
  }
}

Configuration is read from ~/.config/bookpeek/config.toml; command-line options override it. Downloaded Vosk models are stored under ~/.config/bookpeek/models/ so config and models share one directory tree.

bookpeek config new creates that file with the default settings. It will not overwrite an existing config unless bookpeek config new --force is used.

[asr]
engine = "both"
whisper_model = "tiny.en"
vosk_model = "vosk-model-small-en-us-0.15"
device = "auto"
seconds = 30

[extract]
spacy_model = "en_core_web_sm"

[enrich]
enabled = false # set true to always perform web lookup
openlibrary = true
goodreads = true
audnexus = true
# Search order matters: earlier regions win ties and are returned first.
audible_regions = ["us", "uk", "ca", "au"]

Supported Audnexus/Audible regions are us, uk, ca, and au. audible_regions is an ordered priority list, not just a set of enabled regions: if equivalent matches are found in multiple stores, the first region listed wins and its result is returned first. --regions replaces this priority order for one scan, so --regions ca,us gives Canada precedence over the US.

-w / --online may query Open Library, Goodreads, and Audnexus. Goodreads support is installed as a package dependency. Set [enrich].enabled = true to perform that lookup on every scan without passing the flag. Use --offline to override that setting for a single scan. Narrator lookup is performed through Audible catalog search followed by Audnexus metadata lookup when matching is enabled.

CI and PyPI releases

GitHub Actions runs tests, Ruff, and mypy on pushes to main and pull requests. After those checks pass, a change to the package version on main automatically creates the matching GitHub release, which triggers PyPI publishing through Trusted Publishing.

To release a new version:

poetry run python scripts/set_version.py 0.2.0
git add pyproject.toml src/bookpeek/__init__.py
git commit -m "Bump package version"
git push origin main

The repository's PyPI project must have a Trusted Publisher configured for the GitHub repository, workflow .github/workflows/publish.yml, and pypi environment before the first release.

Homebrew releases update brandonscript/homebrew-tap automatically when the HOMEBREW_TAP_TOKEN repository secret is configured. The token needs fine-grained Contents: Read and write access to that tap repository. Without the secret, the Python release still succeeds and CI reports that the tap update was skipped.

Development

poetry run pytest
poetry run ruff check .
poetry run mypy src/

The provider URL smoke test is opt-in because it makes network requests:

BOOKPEEK_LIVE=1 poetry run pytest tests/test_urls_live.py