Speaker Verification

Speaker Verification is a solution developed by Cambridge Cognition, powered by Winterlight Labs' speech analysis platform, designed to detect duplicate participant enrolment in clinical trials. It uses machine learning to extract and evaluate acoustic properties from audio recordings, enabling the identification of individuals who have enrolled more than once. The solution is intended for use in clinical trial management, particularly in CNS trials, and claims a 97%+ accuracy rate in de-duplication detection.

The tool addresses participant duplication by analysing voice characteristics rather than linguistic content, creating unique acoustic fingerprints for each participant and comparing them against all other enrolled participants' pre-randomisation audio data. It can function as a standalone solution or integrate within existing workflows and systems.

Core Capabilities

• Acoustic analysis of pitch, tone, cadence, and other acoustic properties to identify duplicate speakers
• Proprietary acoustic fingerprinting technology that creates unique voice profiles for each participant
• Systematic comparison of a participant's acoustic fingerprint against all other enrolled participants' audio profiles
• Real-time processing that delivers immediate results during the screening process
• Global language support across trial populations
• Full compliance with GDPR and HIPAA regulations

Use Cases in Clinical Trial Management

• Participant duplication detection — identifying when the same individual has enrolled more than once
• At-home assessment authentication — verifying participant identity during remote assessments
• Rater verification and authentication — confirming the identity of raters conducting assessments

Workflow Steps

1. Initiate recording: The rater starts an audio recording at the beginning of the session, capturing all voices present including the participant, rater, and any additional speakers.
2. Transcribe and label: Speakers are identified and transcribed, and subsections within the assessment scale are marked (for example, "Word recall" and "Commands" for ADAS-Cog).
3. Quality check: The transcript and audio are scanned for scale-specific quality indicators to confirm that prompts were read correctly, in the correct sequence, and that appropriate follow-ups were made.
4. Generate report: A quality report is produced for each scale, specifying detected quality indicators and their locations within the assessment.

Duplicate Detection Process

1. Audio capture and analysis: The system analyses existing audio recordings from trial participants, focusing on acoustic characteristics rather than spoken content.
2. Acoustic fingerprinting: Intricate acoustic characteristics including pitch, tone, and cadence are examined to create a unique voice fingerprint for each participant.
3. Systematic comparison: The acoustic fingerprint is compared against the profiles of all other enrolled participants' pre-randomisation audio data.
4. Rapid results and corrective action: When a duplicate is detected, the sponsor and sites are engaged to limit the participant to a single enrolment and take immediate corrective action.

Speaker Verification is positioned within Cambridge Cognition's broader suite of clinical trial technologies, which includes digital cognitive assessments, eCOA, and rater training tools. The solution is built with security and participant consent as foundational requirements, and is designed to reduce costs by catching duplications earlier in the screening process.