Parametrically varying speech adapter length suggests two mechanisms for talker adaptation

Abstract

Listeners are slower at processing mixed-talker speech than single-talker speech. This effect of talker variability is reduced when listeners can reorient their attention to each talker via auditory streaming. However, it remains unknown whether sufficient time for attentional reorientation can make speech processing in a mixed-talker context as efficient as processing single-talker speech. Here, we examined how speech processing efficiency for word identification changes with varying lengths of preceding speech. In single- and mixed-talker conditions, listeners identified target words in isolation or preceded by a carrier vowel (/⋀:/) of parametrically varying durations (300–1500 ms). Word identification was significantly slower in mixed-talker conditions versus single-talker conditions. The response time difference between the mixed- and single-talker conditions significantly decreased as the carrier duration increased from 0 to 600 ms, but carriers longer than 600 ms did not further reduce the additional processing costs in mixed-talker contexts. These results suggest two distinct mechanisms associated with speech processing efficiency: One for rapid adaptation up to 600ms, reflecting attentional orientation via auditory streaming; and another mechanism operating on longer timescales, perhaps reflecting cognitive resource allocation to accommodate the possibility of talker variability. Our ongoing work explores how these two processes operate over natural speech in mixed-talker contexts.