Show HN: Whisper transcribes 70-year-olds more accurately than 20-year-olds

OpenAI's Whisper speech recognition model actually transcribes older speakers (70-year-olds) more accurately than younger speakers (20-year-olds), contrary to common assumptions about age-related speech recognition failures. Testing on 2,760 matched voice clips showed word error rates of 6.53% for people in their twenties, 5.23% for those in their sixties, and 4.67% for those in their seventies. However, a separate finding reveals that when voice systems use fixed silence thresholds for turn-taking, older speakers experience higher rates of being talked over—though this gap narrows significantly with more sophisticated turn-detection models.