Self-hostedWorking build

scribe.

Your audio, transcribed on your own GPU.

An open-source alternative to Otter.ai that runs entirely on your own hardware. Whisper large-v3 transcription with optional speaker labels, wrapped in a web UI, an API and a CLI. No cloud, no accounts, no per-minute billing. Your audio never leaves the machine.

Features

Whisper large-v3

faster-whisper on CTranslate2 with CUDA int8_float16. Fits and flies on an 8 GB consumer GPU.

Speaker diarization

pyannote.audio tells speakers apart. Optional: without it scribe still transcribes and says why labels are missing.

Web UI

Drag and drop a file, watch progress live, read a color-coded transcript per speaker.

API and CLI

Upload, poll and fetch results as JSON, or run one command in a terminal.

Five export formats

TXT, SRT, VTT, JSON and Markdown, from the UI, API or CLI.

Private by design

Everything runs and stays local, including the model cache and the job queue.

Quickstart

git clone https://github.com/jay739/scribe && cd scribe
powershell -ExecutionPolicy Bypass -File scripts/setup.ps1
.venv\Scripts\python -m scribe.server
# open http://localhost:8323

Credit where due

An open-source alternative to Otter.ai. Built independently on faster-whisper and pyannote.audio.