SoundStudio Helper
==================
A small program that runs on this PC and gives the SoundStudio web app tools a browser
cannot run by itself:

  - Stem separation (Demucs): vocals, drums, bass, other
  - Rubber Band time stretch, pitch shift and Fit to Length
  - AI noise reduction (DeepFilterNet, or RNNoise for speech)
  - Lyric / speech timing (Whisper), saved as SRT and LRC too
  - MP3 320, FLAC, M4A and Opus export, and opening any audio format (FFmpeg)
  - Watch folder: drop songs in, get loudness-matched copies and click lists out

It starts by itself when you sign in to Windows (Startup folder shortcut
"SoundStudio Helper"). It only listens on this computer (127.0.0.1:8737) and only answers
the SoundStudio page on jeffapierson.com.

Requirements:  Windows 10 or 11, 64-bit. 8 GB RAM (16 GB or more for stem separation
               of full songs). About 4 GB free disk for the program and AI models, plus
               room for job files. Internet for setup and the first use of each AI model;
               after that everything runs offline. No graphics card needed (runs on the CPU).
Setup:         unzip, then double-click install.cmd. It installs FFmpeg, uv, Node.js (if
               missing), Python 3.11 and the AI tools into %USERPROFILE%\SoundStudioHelper,
               adds a Startup shortcut and starts the helper. Run it again any time to
               repair or update; it never deletes anything.
Status page:   http://127.0.0.1:8737/
Start it:      double-click "Start SoundStudio Helper.cmd"
Stop it:       Task Manager > Details > pythonw.exe (the one running server.py) > End task
Stop autostart: delete the "SoundStudio Helper" shortcut in
               %APPDATA%\Microsoft\Windows\Start Menu\Programs\Startup
Log:           helper.log
Job files:     work\jobs (one folder per job: input and results). Nothing is ever deleted
               automatically; clear old job folders yourself when you want the space back.
Watch folder settings: set them in SoundStudio's Helper tab (saved in config.json).

Installed pieces (in .venv): Python 3.11, PyTorch 2.0.1 CPU, Demucs 4.0.1,
faster-whisper, DeepFilterNet 0.5.6. FFmpeg (with Rubber Band and RNNoise) comes from
the FFmpeg already installed through winget. Models download on first use:
Demucs htdemucs (about 80 MB), Whisper small (about 480 MB), DeepFilterNet3 (about 10 MB).
