python-mindrove-emg

Tools for recording, streaming, replaying, and analyzing MindRove EMG + IMU data.

This project is currently alpha software. It is intended for research workflows around MindRove Wi-Fi EMG devices, Lab Streaming Layer (LSL), XDF recordings, and quick gesture/state separability checks.

Features

• Stream one MindRove device to LSL as EMG and IMU streams.
• Record MindRove EMG + IMU data to compressed .npz files.
• Stream either or both of two MindRove Wi-Fi devices from one GUI.
• Apply optional EMG DC removal and bandpass filtering in the dual-device streamer.
• Replay .npz or .xdf recordings to LSL.
• Analyze event-labeled XDF recordings for state separability.

Installation

For GUI acquisition and streaming:

pip install "python-mindrove-emg[gui]"

For XDF replay support:

pip install "python-mindrove-emg[gui,playback]"

For offline separability analysis:

pip install "python-mindrove-emg[analysis]"

For development from a local checkout:

pip install -e ".[all,dev]"

Command-Line Tools

Single-device LSL streamer:

mindrove-lsl

Single-device recorder:

mindrove-recorder --output-dir recordings --session-name mindrove_session

Dual-device LSL streamer:

mindrove-dual-lsl

Dual-device streamer with custom LSL prefixes:

mindrove-dual-lsl --device-a-prefix MindRove_Left --device-b-prefix MindRove_Right

Dual-device streamer with explicit device IPs:

mindrove-dual-lsl --device-a-ip 192.168.4.1 --device-b-ip 192.168.5.1

Separability analysis for event-labeled XDF recordings:

mindrove-separability C:\path\to\xdf_folder --output-dir analysis_outputs\session001

Data Streams

The LSL tools expose:

• EMG: 8 channels, float32, nominally 500 Hz.
• Combined IMU: 9 channels, accelerometer + gyroscope + magnetometer.
• Optional IMU breakout streams in the dual-device GUI: accelerometer, gyroscope, magnetometer.

Recording And Analysis Notes

The separability analyzer expects XDF files with:

• A continuous MindRove signal stream.
• A marker stream with prompt_onset events.
• Gesture markers encoded like prompt_onset|phase=gesture|trial=001|gesture=hand_open|duration_s=5.000.

It extracts gesture windows, computes time-domain EMG features, and reports leave-one-recording-out accuracy, trial majority-vote accuracy, confusion matrices, pairwise accuracy, centroid distances, and a PCA plot.

Repository Hygiene

Do not commit local recordings, generated analysis outputs, virtual environments, __pycache__, or *.egg-info directories. These are ignored by .gitignore.

License

MIT. See LICENSE.