Contributors are shown a topic from a bank of 5,000, tap record and talk for at least sixty seconds without notes. Scripted readings, clips that stop short of a minute and recordings dominated by silence are sent back. Expect spontaneous monologue with the disfluencies, repairs, fillers and planning pauses intact, in English as spoken by a global base strongest in Southeast Asia, the United States and Europe, five clips per speaker at most, each clip mapped to its topic. The largest open speech corpora (LibriSpeech, Common Voice) are read speech, and models trained on them degrade when people think while talking; this set is for evaluation and fine-tuning of disfluency-aware recognition.