tauri-eeg

Desktop EEG and regulation workspace built with Tauri 2, React, TypeScript, Rust, and a local Python music generation service.

Features

• Local multi-user login backed by SQLite.
• EEG acquisition workspace with realtime waveform display, channel controls, and animated page entry.
• Video, game, and music regulation pages.
• Emotion regulation effect evaluation loop (/effect-evaluation): baseline/post scale pairing, improvement-rate verdicts, history review, and CSV/JSON report export.
• Music page with a compact player, layered prompt builder, generated WAV history, progress display, and file deletion.
• Local Stable Audio 3 Small Music generation through music-service.

Effect Evaluation Loop

The /effect-evaluation page closes the emotion-regulation evaluation loop for a subject:

1. Pick a subject id, target emotion (焦虑 / 抑郁 / 恐惧), regulation method (music or video), and duration.
2. Fill the baseline mental scale (phase=baseline) — reuses the shared scale dialog.
3. Run the regulation on the music/video page while the wizard counts down; the window is a hard floor (leaving early requires a double-confirmed skip that is recorded with the run). An EEG recording is linked automatically when the device is available.
4. Fill the same scale again (phase=post).
5. The backend pairs both records and computes per-dimension improvement (baseline - post) / baseline. All four dimensions (anxiety, worry, mood, energy) are scored so higher means worse, so this single formula applies to every dimension without inversion. The mean improvement rate at or above the 10% threshold (DEFAULT_IMPROVEMENT_THRESHOLD in src-tauri/src/scale_records.rs) marks the emotion as regulated effectively.

Every completed scale submission is persisted to the SQLite scale_records table (in addition to the in-memory status cache), so runs survive restarts. The history tab reviews all completed baseline/post runs, and the export buttons write one run's report (JSON or CSV) or an all-subjects batch summary (CSV) to a user-chosen path through the native save dialog. CSV fields are escaped by hand (RFC 4180, BOM included for Excel); no CSV dependency is used.

Requirements

• Node.js and pnpm.
• Rust toolchain for Tauri.
• Tauri CLI through the project dependency: pnpm tauri ....
• Python package manager uv for music-service.
• Hugging Face account with access accepted for stabilityai/stable-audio-3-small-music.
• NVIDIA CUDA is optional but recommended for music generation.

Install

cd D:\tauri-eeg
pnpm install

Set up the Python music service:

cd D:\tauri-eeg\music-service
uv sync

For RTX 50-series / CUDA 12.8 systems, install the CUDA dependency extra:

cd D:\tauri-eeg\music-service
uv sync --extra cu128
uv run python -c "import torch; print(torch.__version__, torch.version.cuda, torch.cuda.is_available())"

Hugging Face Cache

The Stable Audio model is downloaded through Hugging Face. To keep the model cache in a stable location on Windows:

[Environment]::SetEnvironmentVariable("HF_HOME", "D:\.hf-cache", "User")

Close and reopen PowerShell, your IDE, and the Tauri app after changing the user environment. For the current terminal session only:

$env:HF_HOME="D:\.hf-cache"

Verify login:

cd D:\tauri-eeg\music-service
uv run hf auth whoami

If needed:

uv run huggingface-cli login

Run The App

Development frontend only:

cd D:\tauri-eeg
pnpm dev

Tauri desktop app:

cd D:\tauri-eeg
pnpm tauri dev

Music Generation Flow

The Music page tries to start music-service automatically when generating a track. On first use, the model download and load can take longer than the app startup wait. If the app reports:

Music generation service did not become ready.

start the service manually once:

cd D:\tauri-eeg\music-service
$env:HF_HOME="D:\.hf-cache"
uv run python server.py

Keep that terminal open. In another terminal, check readiness:

curl http://127.0.0.1:8000/health

When the health response is ready, generate again from the Music page. The app reuses the running service at http://127.0.0.1:8000.

Notes:

• flash_attn warnings are expected when Flash Attention is not installed; the service falls back without it.
• on_event is deprecated is a FastAPI deprecation warning and does not block generation.
• WinError 10048 means another service is already using port 8000; stop the old process or reuse it.
• RTX 5090 / RTX 50-series GPUs need CUDA wheels that support sm_120, so use the cu128 extra.

Data Locations

Current Windows paths:

User database:
C:\Users\<you>\AppData\Local\tauri-eeg\users.sqlite3

Generated music WAV files:
C:\Users\<you>\AppData\Roaming\com.tauri-eeg.app\music

Hugging Face model cache, if configured:
D:\.hf-cache\hub

Generated WAV history can be opened from the Music Player history button. Deleting a generated history item from the app also deletes its WAV file, limited to the app music output directory.

Verification

Frontend build:

pnpm build

Frontend tests:

pnpm test

Rust formatting and tests:

cd D:\tauri-eeg\src-tauri
cargo fmt --check
cargo test

Python music-service tests:

cd D:\tauri-eeg\music-service
uv run pytest

Project Structure

src/                 React UI, auth, EEG, and music client code
src-tauri/           Tauri/Rust backend commands, SQLite, file handling
music-service/       FastAPI Stable Audio generation service
docs/superpowers/    Design specs and implementation plans