Make some noise.
Find your signal.
Two black holes spiral together. Their signal rises in pitch.
Can you make a chirp that stands out from the noise?
A chirp in the quiet
Press Record a chirp. Wait for 3, 2, 1, then chirp for 1.5 seconds. Audio stays on your device.
EDUCATIONAL SCORES · χ² AND SG: LOWER IS BETTER
Follow the rising pitch
Does the whole chirp fit?
Signal, or a short glitch?
Hear the universe.
Gravitational waves stretch and squeeze space. Shift their frequencies into our hearing range, and a merger sounds like a chirp.
Be the black holes.
Hum, whistle, or say “whoooop” for 1.5 seconds after the countdown. Follow the curve: a gentle rise, then a faster sweep at the end.
Loud isn’t enough.
As noise increases, a convincing signal needs the right shape. Learn how consistency tests help separate chirps from glitches.
Inside the detector Educational model · how the scores work
A voice-sized analogy for a cosmic search
This is an outreach simulation, not a gravitational-wave analysis pipeline. The target is an accelerating 180–720 Hz sweep, not a physical binary waveform. Each attempt has a three-second countdown followed by a 1.5-second recording. The first countdown second measures background noise; stay quiet until the recording begins. The microphone turns off after each attempt. You can choose a lower or higher register below. Noise is added numerically to the captured audio, not played into the room.
Chirp match: compares the strongest spectral peak in each time frame with the rising target, allowing a small pitch tolerance. It is a time–frequency shape score, not matched-filter SNR.
chisquared: divides the expected chirp power into six equal-power time regions and compares each region’s measured, background-subtracted ridge energy with its expected share. The displayed χ²-like score uses a heuristic noise scale. Lower values mean a more consistent envelope; it is not the Allen frequency-bin χ² or a calibrated reduced χ².
chisquared_sgveto: projects the end of the recording onto short Gaussian-windowed sine/cosine tiles above the target’s final frequency, normalised by an earlier noise recording. Excess tile energy raises the score. This illustrates a sine-Gaussian veto; it is not PyCBC’s implementation or a statistically calibrated veto. Voice harmonics can also raise it: try a soft hum or whistle.
Real searches use noise-weighted matched filtering, frequency-bin χ² and sine-Gaussian checks. Read the PyCBC veto documentation ↗. Thresholds here are chosen for gameplay. A pass is a teaching outcome, not a detection claim.