Quick start¶

1. Install dependencies¶

python scripts\install_local.py

or pip install -r requirements/local.txt

2. Download ASR model (required)¶

python scripts\download_model.py

Script	Size	Required
`download_model.py`	~2.5 GB	Yes (transcription)
`download_summary_llm.py`	~2 GB	Only for local Qwen summary

Cloud summary (DeepSeek, etc.): no model download — only an API key at /settings/summary.

3. Start the web UI¶

start.bat

or sbobina ui / python scripts\restart_ui.py

Open: http://localhost:8501

With Docker (docker compose --profile cpu up), the UI is at http://localhost:8502.

4. Transcribe a file¶

(Optional) Summary settings → DeepSeek or other API key
Sidebar → Generate summary + engine
Upload file → Queue transcription
Track progress on /jobs/{id}; full history at /jobs
Results in data\output\jobs\YYYYMMDD_HHMMSS_filename\

5. CLI¶

sbobina transcribe data\input\myfile.wav -s --summary-provider deepseek

6. Clear history¶

python scripts\clean_output.py

Structure after first use¶

sbobinator/
├── data/
│   ├── .secrets/       ← summary API keys
│   ├── input/
│   └── output/jobs/
├── models/
│   ├── parakeet-tdt-0.6b-v3.nemo
│   └── qwen2.5-3b-instruct/   ← optional
└── .venv/

Common issues¶

Symptom	Solution
`ffmpeg not found`	`winget install Gyan.FFmpeg`
Missing dependencies	`pip install -r requirements/local.txt`
DeepSeek `Connection error`	`pip install truststore` + restart UI
Local summary disabled	RAM < 16 GB or missing GGUF

See Troubleshooting.