Procedure and stimuli

Masking experiments

For the masking experiments, a three-interval forced-choice procedure with adap-tive signal-level adjustment was used to determine masked thresholds. Each trial was 2,9 s long. The masker was presented for the whole duration of the trial.

Temporally centered in the masker, there were three intervals of 300-ms duration separated by 300-ms pauses. Thus the masker was switched on 700 ms before the first interval and switched off 700 ms after the third interval including two 50-ms raised-cosine ramps at on- and offset. The intervals were indicated on the screen in front of the subject. One randomly chosen interval contained the signal. The signal was an So or Sπ sinusoid of 500 Hz. It had a duration of 300 ms including 50 ms raised-cosine ramps at on- and offset. The task of the subject was to in-dicate this signal interval after each trial. Responses were given by pressing the corresponding button (1, 2, or 3) on a computer keyboard. Trial-by-trial feed-back was provided. Signal level was adjusted according to a one-up two-down rule tracking the 70.7% correct response level (Levitt, 1971). The initial step size of the signal level was 8 dB. The step size was halved after each second reversal of the level-adjustment procedure until a step size of 1 dB was reached. With this minimum step size of 1 dB the run continued for another six reversals. The mean over these last six reversals was used as a threshold estimate.

The same noise masker was presented to both ears (diotic presentation). The maskers were generated in the frequency domain using a 2¹⁷-point (0.34-Hz reso-lution) buffer. Their spectrum had a constant nonzero amplitude in the passband regions and a random phase. The lower frequency limit was always set to 30 Hz, the higher to 1 kHz. The noise maskers were transformed to the time domain (inverse FFT) and restricted to the desired length of 127890 samples (2.9s).

For the notched-noise experiment, the notch width was either 0, 50, 200, 400, or 600 Hz. The notch was arithmetically centered at the signal frequency. Stimuli and procedure were similar to those used by Hall et al. (1983). The masker spectrum level was 50dB/Hz. Nitschmann et al. (2009) measured thresholds for NH subjects for this spectrum level.

For the broadband masking experiment, a bandpass-noise masker with no notch was used. In general, the masker spectrum level was 10, 20, 30, 40, or 50 dB/Hz.

For the highest spectrum level, the experimental condition was the same as in

5.2. Methods the notched-noise masking experiment with a notch width of 0 Hz. Thus, to reduce measuring time, the threshold for this spectrum level was taken from the notched-noise experiment.

For five HI subjects (1, 2, 4, 6, and 8), these two masking experiments were also performed with a spectrum level that was adjusted to elicit the same loud-ness as for the NH subjects. For this purpose, loudloud-ness functions of third-octave bands geometrically centered at 125, 250, 500, and 1000Hz were estimated for each HI subject using the method of adaptive categorical loudness scaling (ACALOS, Brand and Hohmann, 2002). From the loudness functions the spectrum level was estimated which produces the same specific loudness at these frequencies as for NH subjects. For intermediate frequencies, the spectrum levels were interpolated from the values for the measured frequencies. Due to the differences in the individ-ual loudness functions the resulting spectrum level was an individindivid-ually different function of frequency. A similar procedure was used in Verhey et al. (2006) to ensure the audibility of all frequency components in an experiment on spectral loudness summation in HI subjects. In the following, the experiments using the same spectrum level as in the NH study will be referred to as the equal-masker-level condition, while those measured at the same specific loudness will be referred to as equal-masker-loudness condition. Note that the same specific loudness does not necessarily imply the same overall loudness since spectral loudness summation is usually smaller for HI than for NH subjects (e.g., Verhey et al., 2006).

Runs with diotic and dichotic stimulus conditions were mixed in the notched-noise and broadband masking experiments. At least three threshold estimates were obtained and averaged for each parameter value and subject. ³ In addition to the masked thresholds, the threshold in quiet was measured for a diotic 500-Hz sinusoid using the same forced-choice procedure as for the masking experiments.

Three criteria had to be fulfilled for a threshold measurement in order to be considered as valid: (i) The standard deviation of each single threshold estimate calculated from six turning points had to be below 3 dB. If the standard deviation of a single threshold estimate was above 3 dB, it was discarded and remeasured.

This applied to a total of three thresholds of the notched-noise experiment. (ii) If the second threshold estimate in a certain masking condition was more than 2 dB below the first and the third was lower than the second or not more than 20% of

3Subject 2 did not finish the experiments for the equal-masker-loudness condition. Since the intraindividual standard deviations of the obtained thresholds are small, the thresholds are nevertheless shown here.

the difference of first minus second higher than the second threshold estimate, this was regarded as a learning effect. In this case the thresholds were remeasured and the first threshold estimates were discarded. This applied to two diotic thresholds of the broadband masking experiment. (iii) In total, three threshold estimates were obtained and averaged for each parameter value and subject. The standard deviation of the mean over these threshold estimates had to be below 3 dB. If it was above 3 dB, more threshold estimates for the respective parameter values were obtained and the previous ones discarded. This did not occur in the experiments of the present study.

Pitch experiment

The ability of the subject to perceive Huggins’ pitch (Cramer and Huggins, 1958) was tested in a way similar to Santurette and Dau (2007): Ten noise intervals, each of 1 s duration including two 50-ms raised-cosine ramps at on- and offset, were presented to the subject at a level of 80 dB SPL. The noise was generated in the frequency domain as described for the masking experiment. In contrast to the masking experiment, a 2¹⁶-point buffer was used because of the shorter stimulus duration. The noise was broadband (30 to 1000 Hz) with random phases.

Noise intervals 1 and 10 contained diotic noise. Noise intervals 2 to 9 contained a major scale in Huggins’ pitch. Huggins’ pitch was introduced by a linear transition of the interaural phase difference from 0 to 2π in a narrow frequency band. The width of the Huggins-pitch band was 16% of its center frequency. Culling et al.

(1998) showed that an interaural-phase transition bandwidth of this size ensures a good perceptibility of the Huggins’ pitch. The major scale was in twelve-tone equal temperament and geometrically centered at 500 Hz, i.e., the center frequency of the lowest and highest Huggins’ pitch were 353.6 and 707.1 Hz, respectively.

Subjects who did not report the percept of a scale or something increasing in pitch after two presentations of the Huggins-pitch scale were given a forced-choice task with two stimuli of different Huggins-pitch center frequency. In this forced-choice task, the center frequencies of the bands with an interaural phase difference other than 0 or 2πwere fixed at 594.6 and 420.4 Hz, i.e., a frequency ratio of√

2:1, geometrically centered at 500Hz. If Huggins’ pitch is perceived, this difference should be clearly audible since the pitch of these two center frequencies is in a frequency region where most HI subjects participating in the study by Santurette and Dau (2007) showed a good frequency discrimination. The spectrum level of

5.2. Methods

0 10 20 30 40 50 60 70 80

90 subject 1

threshold /dB HL

subject 2 subject 3 subject 4

125 500 1k 2k 4k 0

10 20 30 40 50 60 70 80

90 subject 5

threshold /dB HL

frequency /Hz

125 500 1k 2k 4k subject 6

frequency /Hz

125 500 1k 2k 4k subject 7

frequency /Hz

125 500 1k 2k 4k left ear

right ear

subject 8

frequency /Hz

Figure 5.1.: Audiograms for the left (crosses) and right (circles) ears of the eight HI subjects who participated in the study. The audiograms are arranged according to decreasing threshold at 500 Hz measured diotically in a forced-choice procedure.

the noise was 50 dB/Hz. The Huggins-pitch stimuli were presented in a three-interval forced-choice procedure. Each stimulus three-interval had a duration of 600 ms including two 50-ms raised-cosine ramps at on- and offset. The intervals were separated by 300-ms silence intervals. One randomly chosen stimulus interval contained the Huggins-pitch stimulus with the lower pitch, the other two intervals contained the stimulus with the higher pitch. The subject’s task was to indicate the interval with the lower pitch. A run consisted of 21 trials. The number of correct responses was counted. The measurement was stopped if there was a run without a wrong answer or, if no improvement over runs could be observed, after three runs.