Spatial-temporal integration of speech reflections
* Presenting author
In reverberant rooms speech is reflected at boundaries and objects and superimposes with the direct sound, thus creating a complex pattern of temporally delayed, spectrally modified and spatially distributed copies of the direct sound. The fundamental assumption of standard measures and models employed to predict speech intelligibility is that reflections arriving briefly after the direct sound can be integrated and are, hence, useful for speech intelligibility, while reflections arriving later than about 50 to 100 ms after the direct sound are detrimental. This assumption was challenged in a series of experiments within this study by systematically varying the energetic, temporal and binaural properties of direct sound, the reflections, and a stationary noise masker. Speech reception thresholds were measured in normal-hearing listeners. In conditions where either energy or binaural information favor the early components of the room impulse response (RIR), the data confirm that adding reflections with delays beyond a critical time window cannot be perfectly integrated. However, in conditions where the later RIR components are favorable in terms of energy or binaural information, the auditory system appears to ignore the early components and exploit the late components instead. This cannot be modeled by any current speech intelligibility prediction models.