Voice analyzer
live demoLive spectrogram + MPM pitch + CPPS (the clinical-literature clarity measure) — all in-browser, vs. your own baseline
PRD F8 · Phase 2–3
Everything runs in your browser — nothing recorded, nothing uploaded. Under the hood: McLeod Pitch Method (pitchy) with clarity gating, a log-frequency spectrogram, a Praat-style smoothed Cepstral Peak Prominence — the literature's most robust dysphonia correlate, implemented in-browser here likely for the first time — and the eGeMAPS alpha ratio. Consumer mics + AGC make absolute values untrustworthy, so everything is framed vs. YOUR baseline.
🎙️
Hum a comfortable note. Watch the harmonics stack on the spectrogram, set a baseline, then try to come back to it — real-time visual-acoustic biofeedback, the seed of the future listening coach.
Why these four numbers
Pitch + steadiness are the trainable basics. CPPS is the measure the clinical literature keeps landing on for overall voice clarity (published Praat cutoffs discriminate disordered voice at ~94% accuracy on connected speech — but those come from calibrated lab pipelines, which is exactly why we show your trend, never a verdict). The alpha ratio hints at breathiness/brightness. On the log-frequency spectrogram, a healthy hum shows an evenly spaced harmonic comb — watching it blur is the breathiness story told visually. Next upgrades per research: SwiftF0 (ONNX) pitch for noisy rooms, and an AudioWorklet pipeline for sample-aligned CPPS frames.