drama-slicer

Python library and backend-callable API for processing video/audio files into transcript-aware video Slice objects.

Install the published package with uv add drama-slicer or pip install drama-slicer. The distribution name is drama-slicer; the Python import package remains slicer.

Preparation

# macOS
brew install ffmpeg

# Ubuntu/Debian
sudo apt update
sudo apt install ffmpeg

# Python deps
uv sync

Local execution

uv run src/main.py data/input/Popeye_forPresident_512kb.mp4

The command prints the generated manifest path when processing succeeds.

Tests

uv --cache-dir /private/tmp/uv-cache run python -m unittest discover -s tests

The test suite uses real ffprobe, ffmpeg, and WhisperX execution. The end-to-end test verifies that all slices continuously cover the original video duration. Test outputs are grouped under data/output/tests/<video-name>/. Repeated exports replace previous manifest and slice files for the same video name.

Python API

from slicer import slice_video_to_output

manifest_path = slice_video_to_output(
    "data/input/Popeye_forPresident_512kb.mp4",
    output_dir="data/output",
    include_video=True,
)

slice_video_to_output() is the only public Python API. It uses local WhisperX to generate transcript timestamps, builds continuous slices that cover the original video timeline, and writes the result to disk. Other modules, models, and helper functions are internal implementation details.

Output format:

• data/output/<video-name>/*_manifest.json: index of all slices, timestamps, text files, optional video files, and sentence data.
• data/output/<video-name>/*_slice_0000_00000_000-00012_000.txt: transcript lines for each slice.
• data/output/<video-name>/*_slice_0000_00000_000-00012_000.mp4: video slice when include_video=True, which is the default for slice_video_to_output().

Packaging and publishing

The distribution is published to PyPI as drama-slicer, while the Python import package remains slicer.

See docs/PUBLISHING.md for the complete versioning, testing, build, credential, publishing, and installation-verification process.

0.1.2 双语字幕入口

新业务从根包导入 analyze_video、slice_video，默认先确认源片硬字幕风险。英语文本软字幕优先于 Whisper，输出 slicer.slice_set.v2；中英文使用共享 cue 时间，支持位置与词跨度。完整调用、系统依赖、旧入口兼容和检测限制见 字幕契约。