Voice Technology to Identify Opioid Use

Using Voice Technology to Identify Opioid Use in Patients in Treatment for Opioid Use Disorder

This study explored whether changes in a person's voice could help identify opioid use in patients with opioid use disorder (OUD). Current methods for determining whether a patient is intoxicated or in withdrawal often rely on self-reporting and clinical judgment, which can be subjective and inconsistent. Drug tests are logistically challenging to administer and can be costly with repeated use.

The project investigated whether physiological changes associated with opioid use could be detected through speech analysis technology. Researchers evaluated whether machine learning methods could identify voice patterns associated with opioid intoxication or withdrawal.

The primary goal of the study was to assess the accuracy of voice-based biomarkers in identifying opioid use. The study also explored relationships between opioid use and specific speech characteristics.

Study Overview

• Status: Completed
• Conditions: Opioid Use
• Intervention / Treatment: Other: Not applicable- observational study

Detailed Description

This study investigated whether changes in a person's voice could be used to identify opioid use in individuals with opioid use disorder (OUD). The opioid epidemic continues to present significant public health, medical, and social challenges in the United States and globally. Clinicians treating patients with OUD often need to determine whether a patient may be actively using opioids, intoxicated, withdrawing, or responding appropriately to treatment. Current approaches commonly rely on self-reporting, interviews, behavioral observations, urine toxicology testing, and clinical judgment. While these methods can be useful, they may also be subjective, resource-intensive, intermittent, invasive, or difficult to implement frequently in routine care settings.

The purpose of this project was to evaluate whether speech analysis technology could provide a more objective, scalable, and non-invasive approach for monitoring opioid-related physiological changes. Human speech is a complex neuromuscular activity that depends on the coordinated function of the brain, respiratory system, vocal tract, and facial musculature. Opioids can affect cognitive processing, respiratory patterns, motor coordination, reaction time, sedation levels, and muscle control, all of which may influence characteristics of speech production. Prior scientific literature has suggested that physiological and neurological conditions can sometimes produce measurable changes in speech patterns. This project sought to determine whether similar measurable changes could be associated with opioid use.

The study focused specifically on analyzing speech recordings from participants with opioid use disorder. Researchers collected voice samples and applied computational analysis methods to evaluate whether acoustic and temporal speech features could distinguish opioid-related states. The project used signal-processing techniques and machine learning methods to analyze a range of speech characteristics that may reflect physiological effects associated with opioid exposure.

Evaluated speech features included acoustic biomarkers commonly studied in speech analytics research. The project investigated whether combinations of these features could be used to identify patterns associated with opioid intoxication or withdrawal.

A major goal of the study was to assess the feasibility of using speech as a physiological biomarker for opioid use monitoring. Researchers evaluated whether machine learning models could reliably differentiate between opioid-related conditions using speech data alone.

The primary objective of the study was to assess the accuracy and feasibility of voice-based biomarkers for identifying opioid use in individuals with OUD. The study also aimed to better understand the limitations and challenges associated with speech-based impairment detection.

As part of the research effort, the project contributed to the development of internal workflows and analytic infrastructure for handling sensitive speech data. Researchers established preprocessing pipelines for audio ingestion, normalization, feature extraction, labeling, quality control, and model evaluation.

The work generated technical findings regarding the feasibility of speech-based opioid detection and highlighted several scientific and engineering challenges associated with this problem space. These included variability in recording environments, differences between speakers, background noise, individual physiological differences, and the difficulty of isolating opioid-related speech effects from unrelated sources of variation. The study also reinforced the challenges associated with developing generalized machine learning classifiers for complex real-world physiological states using speech data alone.

Although the project explored the potential for objective opioid monitoring through speech analysis, the research did not produce a clinically deployable classifier during the study period. However, the project generated valuable information regarding the limitations, feasibility considerations, and technical barriers associated with speech-based opioid detection approaches. These findings informed future research planning, technology-development decisions, and evaluation strategies for impairment-detection technologies.

Overall, the project contributed to ongoing research efforts exploring non-invasive digital biomarkers for substance-use monitoring. The findings from this work may help guide future investigations into speech analytics, physiological monitoring, and machine learning approaches for identifying substance-related impairment and supporting clinical decision-making in addiction medicine settings.

• Study Type: Observational
• Enrollment (Actual): 41

Contacts and Locations

Study Locations

• United States
  • California
    • Loma Linda, California, United States, 92350
      • Loma Linda University Health, 24951 Circle Drive, Nichol Hall, Room #2042

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

• Sampling Method: Probability Sample

Study Population

The study population is adult (18+) English-speaking men and women currently receiving OUD treatment at the Volpicelli Center, capable of consenting and completing the protocol, and free of severe psychiatric comorbidity or chronic speech-affecting conditions. Because participants are drawn from an active treatment population, they are by definition individuals already engaged in care for OUD rather than treatment-naïve or general-population samples - a relevant consideration when interpreting how the voice biomarker findings might generalize.

Description

Inclusion Criteria:

• Male or female
• At least 18 years old
• Be an active patient in treatment at the Volpicelli Center
• Have a diagnosis of Opioid Use Disorder (OUD)
• Ability to read English
• Able to comprehend and are willing to sign the informed consent form and are able to adhere to the protocol

Exclusion Criteria:

• Severe psychiatric comorbidity
• A chronic medical condition that interferes with speaking (note: Acute conditions that impair speech or hearing will not be considered exclusionary, but testing will be deferred until the temporary condition has been resolved)
• Non-fluency in the study language (English)

Study Plan

How is the study designed?

Number of groups / cohorts

1

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
Patients in treatment for opioid use disorder at the Volpicelli Center.; Prospective longitudinal observational cohort study with repeated measures where each participant completed two visits approximately 30 days apart with repeated speech and clinical measurements. This is a prospective observational study therefore no intervention will be applied.	Other: Not applicable- observational study; Not Applicable - Observational Study

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Sensitivity and Specificity	Sensitivity and Specificity of the machine learning model when classifying voice samples into patient's state based on their speech.	12 months after the enrollment

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
List of specific voice features/categories affected by substances	Identify voice features/categories most affected by the following substances: Buprenorphine, Other Opioid, Opioid Antagonist, Stimulant, Sedative, Cannabinoid.	12 months after enrollment

Collaborators and Investigators

• Sponsor: Tenvos Inc.
• Collaborators: National Institute on Drug Abuse (NIDA); Volpicelli Center

Publications and helpful links

Helpful Links

• GitHub Repository with software code for extracting acoustic features, Montreal Forced Alignment and Audio Transcription (a.k.a. pre-processing pipeline).

Study record dates

Study Major Dates

• Study Start (Actual): December 10, 2024
• Primary Completion (Actual): June 30, 2025
• Study Completion (Actual): January 30, 2026

Study Registration Dates

• First Submitted: May 10, 2026
• First Submitted That Met QC Criteria: May 16, 2026
• First Posted (Actual): May 22, 2026

Study Record Updates

• Last Update Posted (Actual): May 22, 2026
• Last Update Submitted That Met QC Criteria: May 16, 2026
• Last Verified: May 1, 2026

More Information

Terms related to this study

• Keywords: Opioids; Voice Biomarkers
• Other Study ID Numbers: 0612; 1R43DA060696-01 (U.S. NIH Grant/Contract)

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO