VOICE Summit 2019 | VED Talks | Jim Harper | July 24
summary
Voice is framed as a 21st‑century thermometer: a smartphone app collects short samples (hold vowels, read a sentence, or say “Alexa what is the weather”) and extracts prosody, articulatory and source features to build device‑agnostic models. Models claim 80% equal error rate for predicting PHQ‑9 ≥10 from 6 seconds of voice, trained on 15,000+ participants with IRB consent; early subsets (≈4,000) show up to fourfold feature changes tied to suicidal ideation. FDA studies and Harvard collaboration are underway for clinical validation.
AI-written summary · about
chapters (7)
- 0:00Introduction to voice health
- 1:04Thermometer analogy in medicine
- 3:12Vocal biomarkers beyond speech
- 7:24Depression screening via voice
- 9:53Suicide risk detection potential
- 10:54Neuroscience behind voice changes
- 11:59Future of proactive voice assistants
related talks
12mDirection-of-Voice (DoV) Estimation for Intuitive Speech Interaction with Smart Device Ecosystems
01h 01mVocal biomarkers for disease detection
36mVoice Enabling Mobile Apps and Custom Hardware Devices
19m‘The Voice’: New Challenges in a Zero UI world
40mBringing mental health data to doctors
06mSensing to Hear: Speech Enhancement for Mobile Devices Using Acoustic Signals
35mHaving a Voice on Android: How hard can it be?
20mThe Decade Ahead: Two Scientists’ Visions for the Future of Technology