sotto 0.1.0

Private voice transcription server — receive audio, transcribe locally, write clean text.

Sotto

Sotto voce: in a quiet voice. Your words, captured reliably, processed privately.

Sotto is a lightweight Python server that receives audio uploads, transcribes them locally with Whisper, generates a short title and summary via LLM, and writes clean text output to a configurable location.

Designed to run on hardware you control. Pairs with the Sotto iOS app for reliable voice capture.

Quickstart

pip install sotto
sotto init          # creates ~/.config/sotto/config.yaml
sotto start         # starts receiver + worker on port 8377

Configuration

sotto init
# Edit ~/.config/sotto/config.yaml
sotto start

Key settings:

• storage.output_dir — where completed transcripts are written (default: ~/.local/share/sotto)
• pipelines — separate pipelines for private (local LLM) and standard (API) processing
• whisper.model — Whisper model size (default: large-v3)
• whisper.device — cuda or cpu
• auth.tokens — bearer tokens for authenticating uploads from the iOS app

API

Upload audio

POST /upload
Content-Type: multipart/form-data
Authorization: Bearer <token>

file: <audio file>
privacy: "private" | "standard"

Returns: {"uuid": "...", "status": "pending"}

Check job status

GET /jobs/<uuid>
Authorization: Bearer <token>

List jobs

GET /jobs?limit=50&offset=0
Authorization: Bearer <token>

Health check

GET /health

Output

For each completed job, two files are written to output_dir/completed/YYYY/MM/:

• <uuid>.txt — clean transcript
• <uuid>.json — metadata (title, summary, duration, pipeline info)

Prerequisites

• Python 3.11+
• Ollama running separately (for private pipeline)
• Anthropic API key (for standard pipeline)
• CUDA-capable GPU recommended for Whisper transcription

License

MIT — Built by Reed Shea. Open source. Private by design.