transcribe-studio 0.2.3

Local classroom audio transcription with projects, WER evaluation, and pluggable transcript formats

Transcribe Studio

A local, browser-based tool for classroom audio transcription. Organize work by project, split audio into timed chunks, label speakers in free text, and evaluate human transcripts against LLM output (WER + semantic WER).

Built for researchers and annotators who need millisecond timestamps and exportable data — without Label Studio complexity.

Features

• Projects — group recordings by class, session, or study
• Waveform editor — divide audio into chunks (by duration or count), overlap speakers at the same timestamp
• Chunk playback — play one chunk at a time with speed up/down (0.25×–2×, keys , / .)
• Exports — TXT, Markdown, JSON, CSV, SRT, WebVTT
• LLM evaluation — paste or upload hypothesis transcripts; strict + semantic WER
• Pluggable formats — timestamp/speaker lines, JSON segments, plain text (TOML-driven)

Quick start

With uv (recommended)

git clone https://github.com/Mishkat-Quantum-Labs/transcribe-studio.git
cd transcribe-studio
uv venv
uv pip install -e ".[dev]"
uv run transcribe-studio

With pip

pip install transcribe-studio
transcribe-studio

Open http://127.0.0.1:8082

Options:

transcribe-studio start --port 8083
transcribe-studio --port 8083          # same as start
transcribe-studio stop                 # free the port if still running
transcribe-studio status
transcribe-studio start --force        # stop old instance, then start

Windows: command not found?

On some Windows setups (especially Microsoft Store Python), transcribe-studio is installed but not on your PATH. Use any of these instead:

python -m app
python -m app --port 8083

Or install with pipx (adds the command to PATH automatically):

pip install pipx
pipx install transcribe-studio
transcribe-studio

Usage

1. Create a project from the dashboard
2. Upload an MP3/WAV/M4A/OGG/FLAC recording
3. Divide the wave into chunks, then transcribe each segment
4. Use Evaluation to compare your transcript against an LLM upload
5. Export when done

Data is stored under ~/.transcribe-studio/ (override with TRANSCRIBE_STUDIO_DATA).

Environment variables:

Variable	Default	Description
TRANSCRIBE_STUDIO_DATA	~/.transcribe-studio	Data directory
TRANSCRIBE_STUDIO_HOST	127.0.0.1	Bind address
TRANSCRIBE_STUDIO_PORT	8082	Listen port

Supabase: open Settings in the sidebar to add your project URL and anon key.

Deploy on AWS (free tier)

cd infra
cp terraform.tfvars.example terraform.tfvars
terraform init && terraform apply

See infra/README.md. Nginx serves port 80 → app on 8082.

Development

uv pip install -e ".[dev]"
uv run pytest

Publishing

PyPI via uv (recommended)

uv build
uv publish   # uses UV_PUBLISH_TOKEN or prompts for PyPI credentials

PyPI via pip/twine

pip install build twine
python -m build
twine upload dist/*

GitHub release

git tag v0.2.0
git push origin v0.2.0
gh release create v0.2.0 dist/*

Configuration

Evaluation and transcript import settings ship inside the package:

• app/config/evaluation.toml
• app/config/transcript_formats.toml
• app/config/languages/en.toml

License

MIT — see LICENSE.

Contributing

Issues and PRs welcome at github.com/Mishkat-Quantum-Labs/transcribe-studio.