PEEC: The Protected Entities Ethics Checklist for Collecting Speech Data From Vulnerable Clinical Populations

Area of Science:

• Speech and Language Technology
• Clinical Research Ethics
• Artificial Intelligence

Background:

• Advancements in automatic speech recognition (ASR) and natural language processing offer clinical potential but are limited to high-resource settings.
• Ethical considerations at the intersection of clinical vulnerability, linguistic diversity, and speech technology are often overlooked.
• Data collection practices directly impact the fairness and bias of ASR systems, linking participant protection to algorithmic justice.

Purpose of the Study:

• Introduce the Protected Entities Ethics Checklist (PEEC), a framework for ethical speech and language data collection from protected populations.
• Address ethical challenges in collecting data from vulnerable groups for ASR and natural language processing applications.
• Establish ethical data collection as a prerequisite for developing fair and equitable AI systems in clinical speech research.

Main Methods:

• Developed the Protected Entities Ethics Checklist (PEEC) framework.
• Structured the PEEC around three core domains: participant protection and consent, data collection standards, and compliance implementation.
• Incorporated population-specific guidance for consent, data protection, quality assurance, and technical considerations for ASR.

Main Results:

• The PEEC provides structured guidance for ethical research with diverse protected entities, including children, elderly adults, individuals with communication disorders, and marginalized communities.
• Offers population-specific consent mechanisms and enhanced data protection measures.
• Includes systematic quality assurance and technical guidance for ASR applications, adaptable to various research contexts.

Conclusions:

• Ethical participant treatment is fundamental to achieving algorithmic fairness in speech technology.
• Procedural justice in data collection is essential for creating fair AI systems.
• Equitable and respectful data collection practices lay the foundation for ASR systems that perform fairly across diverse populations.