Brain Recordings
The Recordings page (/recordings) is your central hub for capturing and cleaning brain signals. Connect an EEG device, record live data, check signal quality in real-time, then denoise your recording with ZUNA β all in one flow.
Connect Device β Record EEG β Check Signal Quality β Stop β Denoise with ZUNA β Compare Before/After β Download
Three Recording Modes
The page offers three modes, selectable via tabs at the top:
π§ Live Recording
Connect your EEG headset and capture brain activity in real-time.
- Connect β Choose your device (Neurosity Crown or Muse) and click Connect. The Muse uses Web Bluetooth; Neurosity uses OAuth.
- Record β Hit Start Recording. Youβll see a live waveform and a signal quality panel showing per-channel contact quality.
- Review β Stop recording when you have enough data (minimum 6 seconds for ZUNA).
- Denoise β Click βDenoise Recordingβ to send your data to the ZUNA GPU for artifact removal.
Record at least 6 seconds β ZUNA processes in 5-second epochs and needs one full epoch plus a small buffer.
π Upload CSV
Upload a previously recorded EEG file in CSV format. The page auto-detects whether the file came from a Neurosity Crown or Muse based on channel names in the header row.
- Supported formats: CSV with channel columns (e.g.,
CP3,C3,F5,PO3,PO4,F6,C4,CP4 for Crown)
- Values should be in microvolts (Β΅V)
- Optional timestamp column
π§ͺ Experiments
Run cognitive experiments (Stroop, Oddball, P300, etc.) from the built-in experiment library. An experimental panel lets you simultaneously record EEG during the experiment:
- Choose an experiment from the dropdown
- Expand the βRecord EEG during experimentβ panel
- Connect your device and start recording
- Run the experiment β brain activity is captured alongside
- Stop and denoise when the experiment completes
The experiment recording feature is marked as Experimental. It records EEG in parallel with the jsPsych experiment but does not yet embed event markers into the EEG timeline.
Live Signal Quality
During active recording, a Signal Quality panel appears showing:
- Bar chart β Standard deviation per channel. Green (1.5β10) = great, Yellow (10β15) = good, Red (>15) = poor contact
- Brain montage β Top-down 2D head map with electrode positions colored by quality
This helps you adjust headset fit before committing to a long recording. Once most channels are green, your data will be high quality.
ZUNA AI Denoising
After recording or uploading, the βClean with ZUNA AIβ step sends your EEG data to the ZUNA foundation model running on an NVIDIA A100 GPU.
What ZUNA Removes
| Artifact | Example |
|---|
| Eye blinks | Frontal voltage spikes from blinking |
| Muscle activity | EMG from jaw clenching, forehead tension |
| Environmental noise | 50/60 Hz line noise, cable movement |
| Drift | Slow baseline wander from electrode impedance changes |
What ZUNA Preserves
Neural oscillations (alpha, beta, theta, delta, gamma), event-related potentials (P300, N170), and other brain-generated signals.
GPU Auto-Wake
The GPU VM auto-sleeps after 30 minutes of inactivity to minimize costs. When you trigger denoising:
- If the GPU is ready β processing starts immediately (~5β15s depending on recording length)
- If the GPU is sleeping β the page automatically wakes it and retries every 15 seconds (up to 6 attempts). Youβll see status updates like βAuto-retrying in 15sβ¦ (attempt 2/6)β
- The GPU status indicator at the top of the page shows real-time availability (green = ready, yellow = starting, gray = sleeping)
You can also manually wake the GPU by clicking βWake GPUβ in the status bar.
Before / After Comparison
After denoising completes, the page shows a side-by-side comparison:
| Left Panel | Right Panel |
|---|
| Before β Raw Signal | After β ZUNA Denoised |
| Your original recording with artifacts | Cleaned signal with artifacts removed |
Both panels use the same scale (Β±50 Β΅V, adjustable) and time axis so you can directly compare waveform quality. The raw signal is captured at the moment you click Denoise, ensuring an accurate snapshot.
After Denoising
Once denoising is complete, you have several options:
- π₯ Download Denoised CSV β Export the cleaned signal as a CSV file for offline analysis in MATLAB, Python, or other tools
- π View in Analysis β Navigate to the
/analysis page for deeper inspection, power spectral density, and more
- Re-record β Start a new recording and denoise again
Result Summary
The results card shows key metadata:
| Field | Description |
|---|
| Input | Number of channels and duration of your recording |
| Output | Channels and duration after processing |
| Channels | List of EEG channel names in the output |
| Sample Rate | Output sampling frequency (256 Hz) |
Supported Devices
| Device | Channels | Connection |
|---|
| Neurosity Crown | 8 (CP3, C3, F5, PO3, PO4, F6, C4, CP4) | Bluetooth + OAuth |
| Muse 2 / Muse S | 4 (TP9, AF7, AF8, TP10) | Web Bluetooth |
Both devices stream at 256 Hz. See Brain Interfaces for detailed setup instructions.
End-to-End Flow
ββββββββββββββββ ββββββββββββββββ ββββββββββββββββ ββββββββββββββββ
β EEG Device ββββββΆβ /recordings ββββββΆβ Backend API ββββββΆβ ZUNA GPU β
β Crown/Muse β β Live viewer β β Auto-wake β β A100 80GB β
ββββββββββββββββ β Sig quality β β /eeg-models β β Denoise β
ββββββββ¬ββββββββ ββββββββββββββββ ββββββββ¬ββββββββ
β β
βΌ βΌ
ββββββββββββββββ ββββββββββββββββ
β Before/Afterββββββββββββββββββββββββββββ Cleaned EEG β
β Comparison β β JSON/CSV β
ββββββββ¬ββββββββ ββββββββββββββββ
β
ββββββββΌββββββββ
β Download / β
β Analysis β
ββββββββββββββββ
For programmatic access to ZUNA (without the UI), see the Model APIs documentation.