Detection Model
How to use: Place your phone call on speaker mode, hold your device near the computer microphone, then press Start Monitoring. Results update every 5 seconds. Monitoring is disabled on this deployment — the microphone is never opened here.
Works best in a quiet environment. Background noise can reduce accuracy. All analysis runs server-side — no API keys or internet required.
Accent-aware: Prosodic checks (pitch, rhythm, pause patterns) are automatically down-weighted and bias-corrected for non-native and accented speech. Physiological checks (micro-jitter, shimmer, inter-harmonic noise) are accent-invariant.
Demo. This page shows the tool's interface and how it scores what it finds. Live analysis is not running for you yet.
Live access is by invitation. Verify an approved email to run this page against your own target, or ask for access.
How it works
Browser captures your microphone in real time using the Web Audio API. Audio is processed locally — nothing is streamed to external servers.
Every 5 seconds: 22 checks including MFCC smoothness, CPP, HNR, micro-jitter, inter-harmonic noise, glottal shimmer, formant trajectory, prosodic emphasis, sub-band kurtosis, and more. Prosodic checks are tagged accent-sensitive.
Physiological checks (jitter, shimmer, HNR, inter-harmonic noise) are accent-invariant — real vocal-fold physics TTS cannot fake. If only prosodic/rhythm checks fire with no physio signals, the engine auto-applies an accent-bias correction and labels it.
Scoring happens in /api/live-call-chunk on the server, which is not enabled on this deployment — no microphone is opened and no audio is captured here.