speechscribe 0.1.0

A small faster-whisper command-line transcriber for local audio files.

speechscribe

A small command-line wrapper around faster-whisper for local audio transcription.

It packages your original script as a pip-installable command:

speechscribe audio.wav

Package name

This project is configured to publish as:

name = "speechscribe"

The installed command and import package are also speechscribe.

Install locally while testing

From this directory:

python3 -m pip install -e .
speechscribe --help

Usage

Transcribe with the default model (medium) and default compute type (int8):

speechscribe audio.wav

Use --input instead of the positional argument:

speechscribe --input audio.wav

Choose a model:

speechscribe audio.wav --model small
speechscribe audio.wav --model large-v3

Choose a compute type:

speechscribe audio.wav --compute-type float16

Write to a custom output file:

speechscribe audio.wav --output transcript.txt

Print timestamped segments as they are processed:

speechscribe audio.wav --verbose

Copy the finished transcript to the macOS clipboard with pbcopy:

speechscribe audio.wav --copy

Skip the delete-original prompt:

speechscribe audio.wav --nodelete

Use the DokuWiki output mode from the original script:

speechscribe audio.wav --format dokuwiki

By default, output is written beside the audio file as:

YYYY_MM_DD_inputfilename.txt

Publish to PyPI

This repository includes a helper script named pip-publish.

Recommended environment-variable token setup:

export TWINE_USERNAME=__token__
export TWINE_PASSWORD=pypi-your-token-here

Then run:

chmod +x pip-publish
./pip-publish

Optional TestPyPI run:

./pip-publish --testpypi

Dry run that builds and checks distributions without uploading:

./pip-publish --dry-run

The script will:

1. install or upgrade build and twine, unless --no-install-deps is passed;
2. clean previous dist/, build/, and *.egg-info output, unless --skip-clean is passed;
3. build a source distribution and wheel;
4. run twine check dist/*;
5. upload with twine upload dist/*.

See docs/PUBLISHING.md for more options.

Notes

• faster-whisper requires Python 3.9 or newer.
• The first run may download the selected Whisper model.
• The --copy option is intentionally macOS-focused because it uses pbcopy, matching the original script.

License

MIT