Random audio-book segments make effective continuous speech maskers

Evidence

A total of 200 masker samples for each interfering talker were created by randomly cutting ongoing speech segments of 3.5-s duration from the preprocessed audio book signals. This length ensured that all target sentences were entirely masked by interfering speech. The onsets and offsets of the segments were shaped by 200-ms raised-cosine ramps. — Dijk, Pim van, Physiology, Psychoacoustics and…, 1628