Effect of priming on energetic and informational masking in a same-different task

Abstract

Objectives: The primary goal of this study was to investigate how speech perception is altered by the provision of a preview or "prime" of a sample of speech just before it is presented in masking. A same-different test paradigm was developed which enabled the effect of priming to be measured with energetic maskers in addition to those that most likely produced both energetic and informational masking. Using this paradigm, the benefit of priming in overcoming energetic and informational masking was compared.

Design: Twenty-four normal-hearing subjects listened to nonsense sentences presented in a background of competing speech (two-talker babble) or one of two types of speech-shaped noise. Both target and masker were presented via loudspeaker directly in front of the listeners. In the baseline condition, the listeners were then shown a sentence on a computer screen that either matched the auditory target sentence exactly or contained a replacement for one of the three target key words. Their task was to judge whether the printed sentence matched the auditory target and respond via computer keyboard. In the first experimental condition, the printed sentence preceded rather than followed the auditory presentation (the priming condition). In the second experimental condition, the perception of spatial separation was created between target and masker by presenting the masker from two loudspeakers (front and 60° to the right) and imposing a 4-msec delay in the masker coming from the front loudspeaker. This resulted in the target being heard from the front while, because of the precedence effect, the masker was heard well to the right (the spatial condition). In a third experimental condition, spatial separation and priming were combined. A total of five signal-to-noise ratios were tested for each masker.

Results: The competing speech masker produced more masking than noise, consistent with previous findings. For the competing speech masker, the signal-to-noise ratio for 80% correct performance was approximately 6.7 dB lower when the listeners read the sentences first (the priming condition) than in the baseline condition. This priming effect was similar to the improvement obtained when the target and masker were separated spatially. Significant priming effects were also observed with speech-shaped noise maskers, and when there was perceived spatial separation between target and masker, conditions in which informational masking was believed to have been minimal. There seemed to be an additive effect of spatial separation and priming in the two-talker babble condition.

Conclusions: (1) Priming was effective in improving speech perception in all conditions, including those consisting of primarily energetic masking. (2) It is not clear how much benefit from priming could be attributed to release from informational masking. (3) Performance on the same-different task was linearly related to performance on an open-set speech recognition task using the same target and masker.

Fig. 1

Long-term spectra of the target (offset +20 dB to facilitate comparison), the female two-talker speech babble (TTB) and the two noise maskers (DSN and SSN) created by digitally signal processing the TTB masker.

Fig. 2

Experimental apparatus in the sound attenuated booth, with loudspeakers at 0° and 60° relative to the listener’s head, a computer monitor (well below ear level) for viewing the primes, and a keyboard for typing responses.

Fig. 3

Group mean-percent correct scores as a function of SNR for the non-spatial primed (F-Fp; closed diamonds), spatial control (F-RFc; open squares), and non-spatial control (F-Fc; open circles) conditions with the two-talker babble (Fig. 3a. TTB; n=6), speech-spectrum noise (Fig. 3b. SSN; n=6), and dynamic speech-spectrum noise (Fig. 3c. DSN; n=6). Arrows indicate horizontal shift at the 80% correct point. Error bars represent ±1 standard error of the mean.

Fig. 4

Group mean-percent correct scores (n=6) as a function of SNR for the spatial primed (F-RFp; closed diamonds) condition, along with additional data for the spatial control (F-RFc; open squares), and non-spatial control (F-Fc; open circles) conditions with the two-talker babble (TTB2 is the same masker used in Fig. 3a with a different group of subjects). The non-spatial primed data from Figure 3a are plotted here for comparison (dashed line). Arrow indicates horizontal shift at the 80% correct point. Error bars represent ±1 standard error of the mean.

Fig. 5

Signal detection data. Group mean (n=6 for each listening condition) hit [P(“Different” | foil)] and false alarm [P(“Different” | non-foil)] rates are plotted for all listening conditions and for each SNR. Closed black symbols = TTB masker, closed gray symbols = SSN masker, open gray symbols = DSN masker, open black symbols = TTB2 masker (same masker as TTB, but data are from a second group of 6 subjects).

Fig. 6

Group mean-percent correct scores for conventional open-set speech recognition (data from Freyman et al. 2007; n=12) plotted as a function of group mean-percent correct scores for the same-different task with the two control conditions F-RFc (closed circles) and F-Fc (open circles); (n=12). The same target sentences in the presence of the same two-talker masker were used in both sets of data. Interpolation of current data to obtain points for comparison at common SNRs resulted in points that are exactly half way between the tested SNRs.