Vocal-biomarker research (e.g. Cummins et al. 2015; Low et al. 2020) repeatedly finds that depressed speech tends toward flatter prosody: reduced pitch (F0) variability, slower speaking rate, more/longer pauses, and reduced loudness variability ("flat affect"). This simulator turns four such features into one illustrative composite score:
d_pitch = clamp((ref_pitch - pitch) / ref_pitch , -1, 1)
d_rate = clamp((ref_rate - rate) / ref_rate , -1, 1)
d_pause = clamp((pause - ref_pause) / (1-ref_pause), -1, 1)
d_energy = clamp((ref_energy - energy) / ref_energy, -1, 1)
z = 1.1·d_pitch + 0.9·d_rate + 1.3·d_pause + 0.8·d_energy
risk = 1 / (1 + e^-(1.6·z - 1.5)) ← logistic squashing
- Pitch variability — standard deviation of fundamental frequency (F0) in semitones over an utterance; monotone speech scores low.
- Speech rate — syllables per second; psychomotor slowing lowers this.
- Pause ratio — fraction of the utterance spent silent; longer/more frequent pauses raise it.
- Loudness variability — how much vocal energy swings within an utterance; flat affect reduces it.
- The reference values (pitch 4.0 st, rate 4.3 syll/s, pause 12%, energy 5.0) approximate a typical non-depressed speaker; the weights rank pause behaviour and pitch flattening as the strongest contributors, matching the literature's feature-importance ordering.
Real-world relevance: this is the same feature family (prosodic/acoustic) used by research-stage passive voice-analysis tools that flag depression/anxiety risk from short speech samples for clinician follow-up — never as a standalone diagnosis.