Show HN: Whisper transcribes 70-year-olds more accurately than 20-year-olds
Summary
A research-style analysis of Whisper's age-related ASR performance finds that transcription accuracy does not degrade for older speakers in the tested dataset, with turn-taking challenges more pronounced under fixed-threshold silencing but mitigated by semantic turn models. The piece compares Whisper, wav2vec2, and Pipecat smart-turn, discusses dataset biases and replication steps, and outlines practical implications for voice-enabled systems.