Spoken Language Analysis in Aging Research: The Validity of AI-Generated Speech to Text Using OpenAI's Whisper

Area of Science:

• Gerontology
• Computational Linguistics
• Psychology

Background:

• Analyzing older adults' speech offers insights into cognitive, affective, and social aging.
• Traditional speech transcription is human-dependent, time-consuming, and resource-intensive.
• AI advancements enable automated speech-to-text (AST) as a potential replacement.

Purpose of the Study:

• To evaluate the accuracy of AI-based AST tools (Whisper, otter.ai) against human-corrected transcripts in aging research.
• To assess the impact of filler word tagging on language analysis validity.
• To determine the readiness of AST for psychological language research.

Main Methods:

• Compared Whisper and otter.ai AST outputs with human-corrected transcripts from 238 older adults.
• Utilized Linguistic Inquiry and Word Count (LIWC) to analyze language features.
• Assessed the effect of manual filler word annotation on LIWC analysis.

Main Results:

• AI-derived LIWC features showed very high convergence with human-derived features (average r = 0.98).
• Manual tagging of filler words minimally affected LIWC validity, except for 'filler words' and 'netspeak' categories.
• High accuracy supports the utility of AI transcription for aging studies.

Conclusions:

• OpenAI's Whisper and otter.ai are effective tools for language analysis in aging research.
• State-of-the-art AI-powered AST is suitable for psychological language research.
• Automated transcription significantly enhances efficiency in analyzing older adults' speech.