realtime-audio

Live microphone audio injected into a spiking population, visualised in NengoGUI.

Description

A small NengoGUI-first model: a nengo.Node reads the most recent ~10 ms of microphone audio (captured via sounddevice / PortAudio), resamples it to a fixed AUDIO_DIMS-length window, and routes the window into a spiking ensemble. The mic samples are written to a ring buffer in the PortAudio callback thread and the node always returns the latest window, so the display stays live whether the simulation runs faster or slower than real time.

Cross-platform — the sounddevice pip wheels bundle the PortAudio binaries on macOS and Windows, so no OS-specific setup is needed beyond granting your terminal microphone permission.

Install

python -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate
pip install -r requirements.txt

Run

nengo realtime_audio.py

In the browser GUI, right-click the audio neurons ensemble to add a value plot and a spike raster, press play, and speak into the mic. The shipped realtime_audio.py.cfg already lays out a value plot and raster in the usual spots.

Tweak AUDIO_DIMS, WINDOW_MS, and GAIN at the top of realtime_audio.py.

To sanity-check the microphone alone (no Nengo / GUI):

python mic_input.py

OS notes

The only OS-specific aspect is microphone permission:

macOS will prompt to allow microphone access for your terminal app the first time you run it (System Settings → Privacy & Security → Microphone).
Windows: ensure Settings → Privacy → Microphone allows desktop apps.
Linux: typically just works once the user is in the audio group; otherwise grant ALSA / PulseAudio access through your distro's normal mechanisms.