SS07: PSYCHOACOUSTICS, SPEECH PERCEPTION AND AUDITORY SIGNAL PROCESSING

MODELLING SPEECH INTELLIGIBILITY IN STEADY AND FLUCTUATING NOISE FOR NORMAL AND HEARING IMPAIRED LISTENERS

Thomas Brand

Carl von Ossietzky Universität Oldenburg, Germany

One possibility for modelling the influence of fluctuating noises on speech intelligibility is the use of short-term extensions of the Speech-Intelligibility-Index (SII). Different short-term extensions are compared that comprise an increasing deviation from the original SII. Even though these approaches differ considerably in complexity, their results are very close to each other.

An important factor is speech context that is used by listeners to replenish dropouts due to peeks of the fluctuating noise. This can be done much more efficiently for highly predictable speech than for speech with low predictability. This can be described using a two-stage context model. In the first stage, the intelligibility of a single word is modelled based on the intelligibility of statistically independent sub-word units. The second stage combines the intelligibilities of the single words in order to calculate the intelligibility of the sentence. The model is based on SRT measurements in quiet and in steady-state noise in order to calculate the SRT in fluctuating noise.