Detection of Alcohol Intoxication Using Voice Features: A Controlled Laboratory Study.

Area of Science:

• Biomedical Engineering
• Speech Processing
• Toxicology

Background:

• Mobile devices and smart speakers offer potential for remote identification of alcohol intoxication through voice alterations.
• Existing data supporting voice analysis for detecting alcohol intoxication in the English language is limited.
• Just-in-time interventions for alcohol use could be facilitated by accurate, remote detection methods.

Purpose of the Study:

• To evaluate the efficacy of spectrographic voice features in identifying alcohol intoxication in English speakers.
• To determine the accuracy of voice analysis models in detecting alcohol intoxication compared to breath alcohol concentration.

Main Methods:

• A controlled laboratory study involving 18 participants (72% male, ages 21-62) who consumed alcohol.
• Voice samples were collected hourly for up to 7 hours post-alcohol consumption after reading a tongue twister.
• Support vector machine models were developed to detect alcohol intoxication (breath alcohol concentration > 0.08%) using spectrographic voice features.

Main Results:

• Alcohol intoxication was predicted with high accuracy (98%, 95% CI [97.1, 98.6]).
• The models demonstrated excellent performance with mean sensitivity of 0.98, specificity of 0.97, positive predictive value of 0.97, and negative predictive value of 0.98.

Conclusions:

• Brief English voice segments, analyzed via spectrographic signatures, effectively identified alcohol intoxication in a controlled setting.
• The findings suggest voice analysis is a promising tool for detecting alcohol intoxication.
• Larger studies with diverse voice samples are recommended to validate and enhance these voice analysis models.