The combined effects of reverberation and nonstationary noise on sentence intelligibility
Iso-STI contours for listening conditions that include a combination of reverberation and stationary noise. The dotted curve represents the .33 iso-STI contour. The data points are measurement results for normal-hearing listeners by Duquesnoy and Plomp (1980), representing the SNR at the SRT, i.e., the level at which 50% of the sentences could be correctly reproduced, for various reverberation times . Redrawn, from Houtgast et al. (1980).
The effect of simulated reverberation on the temporal waveforms of three different types of nonstationary maskers. To be able to compare the masker types in terms of visible waveform modulations, only the modulations for one representative octave band (around ) are shown. Details on the masker types and the reverberation procedure can be found in the Sec. III.
ESII as a function of speech-to-noise ratio (SNR) for three different types of nonstationary maskers with reverberation time as a parameter. ESII values were calculated for each reverberant masker and fitted to the three-parameter asymmetric logistic function reduces to a symmetrical sigmoid for . For stationary noise, ESII is simply the linear function , see ANSI (1997). Details on the masker types and the reverberation procedure can be found in the Sec. III.
Overview of the proposed prediction method. The effect of reverberation on speech is described by the STI method, while the effect of reverberation on the masker is also evaluated separately. Subsequently, the ESII is applied to determine the combined effects of both reverberation and masking noise on sentence intelligibility. The example below applies to a reverberation time of and the Plomp-two band masker type.
SRTs as a function of reverberation, for three combinations of speech corpus (Plomp and VU98) and nonstationary masker type (two band and MLS). Predictions, represented by curves, were determined by applying the STI method with 18 modulations bands, as suggested by Van Wijngaarden and Houtgast (2004). For each speech corpus (Plomp or VU98), the STI at threshold was chosen to give an optimal fit (least squares) between measurements and predictions. The leftmost panel also displays results from SRT measurements in stationary noise, taken from Duquesnoy and Plomp (1980).
Scatter plot of the observed SRTs and the predicted SRTs, for all combinations of speech corpus, masker type, and reverberation.
Group means and standard deviations for speech reception thresholds (SRT, in dB SNR) in ten reverberant conditions, for three combinations of speech corpus (Plomp and VU98) and nonstationary masker type (two band and MLS).
Article metrics loading...
Full text loading...