Spectral tilt change in stop consonant perception by listeners with hearing impairment

Abstract

Purpose: To evaluate how perceptual importance of spectral tilt is altered when formant information is degraded by sensorineural hearing loss.

Method: Eighteen listeners with mild to moderate hearing impairment (HI listeners) and 20-23 listeners with normal hearing (NH listeners) identified synthesized stimuli that varied in second formant (F(2)) frequency and spectral tilt. Experiments 1 and 2 examined utterance-initial stops (/ba/ and /da/), and Experiments 3 and 4 examined medial stops (/aba/ and /ada/). Spectral tilt was manipulated at either consonant onset (Experiments 1 and 3), vowels (Experiments 2 and 4), or both (Experiment 5).

Results: Regression analyses revealed that HI listeners weighted F(2) substantially less than NH listeners. There was no difference in absolute tilt weights between groups. However, HI listeners emphasized tilt as much as F(2) for medial stops. NH listeners weighted tilt primarily when F(2) was ambiguous, whereas HI listeners weighted tilt significantly more than NH listeners on unambiguous F(2) endpoints.

Conclusions: Attenuating changes in spectral tilt can be as deleterious as taking away F(2) information for HI listeners. Recordings through a wide dynamic range compression hearing aid show compromised changes in spectral tilt, compressed in range by up to 50%.

Figure 1

Short-term spectra representing change in tilt for stimuli in Experiment 1 in which different consonant onset tilts converged to a −6 dB/oct. vowel tilt. This example represents stimuli with an F₂-onset frequency of 1400 Hz. Short-term spectra show the first (left) and fourth (right) pitch pulse (about 10 ms each) and were analyzed without pre-emphasis using a 256-point FFT with a 50% Hamming window overlap. Consonant onset tilts shallower than vowel tilt (top) are expected to result in more alveolar responses and consonant onset tilts steeper than vowel tilt (bottom) are expected to result in more labial responses.

Figure 2

Identification rates for mean data of HI and NH listeners in Experiment 1 are plotted in the left and right panels, respectively. Probability of responding /da/ as a function of F₂-onset frequency is plotted separately for each consonant onset tilt. Circles, asterisks, squares, x’s, and triangles represent mean data for consonant onset tilts of −12, −9, −6, −3, and 0 dB/oct., respectively. Maximum-likelihood fits of identification rates are displayed for mean data at each consonant onset tilt as different lines (see legend).

Figure 3

Standardized linear regression weights for tilt at each F₂ frequency for HI and NH listeners (filled squares and open circles, respectively) in Experiment 1. Error bars represent standard error of the estimate. Bonferroni-adjusted levels of statistical significance between groups at each frequency are indicated by asterisks (***p ≤ 0.001; ** p ≤ 0.01).

Figure 4

The 1/3 octave band long-term average speech spectra (LTASS) of consonant onsets (t = 0–30ms) for Experiments 1 and 2 (dotted and solid lines, respectively) plotted against mean thresholds in dB SPL of HI listeners (dashed line). Error bars represent standard error of the mean. Note that LTASS of onsets are essentially identical across the two experiments.

Figure 5

Short-term spectra representing change in tilt for stimuli in Experiment 2 in which different vowel tilts diverged from a common −6 dB/oct. onset tilt. This example represents stimuli with an F₂-onset frequency of 1400 Hz. Short-term spectra are represented as in Figure 1. When consonant onset tilt is steeper than vowel tilt (top) more labial responses are expected, and when consonant onset tilt is shallower than vowel tilt (bottom) more alveolar responses are expected.

Figure 6

Identification rates for mean data of HI and NH listeners in Experiment 2 are plotted in the left and right panels, respectively. Probability of responding /da/ as a function of F₂-onset frequency is plotted separately for each vowel tilt. Circles, asterisks, squares, x’s, and triangles represent mean data for vowel tilts of −12, −9, −6, −3, and 0 dB/oct., respectively. Maximum-likelihood fits of identification rates are displayed for mean data at each vowel tilt as different lines (see legend).

Figure 7

Standardized linear regression weights for tilt at each F₂ frequency for HI and NH listeners (filled squares and open circles, respectively) in Experiment 2. Error bars represent standard error of the estimate. Bonferroni-adjusted levels of statistical significance between groups at each frequency are indicated by asterisks (*** p ≤ 0.001; ** p ≤ 0.01; * p ≤ 0.05).

Figure 8

Identification rates for mean data from HI and NH listeners in Experiment 3 are plotted in the left and right panels, respectively. Probability of responding /ada/ as a function of F₂-onset frequency is plotted separately for each consonant onset tilt. Circles, asterisks, squares, x’s, and triangles represent mean data for onset tilts of −12, −9, −6, −3, and 0 dB/oct., respectively. Maximum-likelihood fits of identification rates are displayed for mean data at each consonant onset tilt as different lines (see legend).

Figure 9

Standardized linear regression weights for tilt at each F₂ frequency for HI and NH listeners (filled squares and open circles, respectively) in Experiment 3. Error bars represent standard error of the estimate. Bonferroni-adjusted levels of statistical significance between groups at each frequency are indicated by asterisks (*** p ≤ 0.001; ** p ≤ 0.01).

Figure 10

Identification rates for mean data of HI and NH listeners in Experiment 4 are plotted in the left and right panels, respectively. Probability of responding /ada/ as a function of F₂- onset frequency is plotted separately for each vowel tilt. Circles, asterisks, squares, x’s, and triangles represent mean data for vowel tilts of −12, −9, −6, −3, and 0 dB/oct., respectively. Maximum-likelihood fits of identification rates are displayed for mean data at each vowel tilt as different lines (see legend).

Figure 11

Standardized linear regression weights for tilt at each F₂ frequency for HI and NH listeners (filled squares and open circles, respectively) in Experiment 4. Error bars represent standard error of the estimate. Bonferroni-adjusted levels of statistical significance between groups at each frequency are indicated by asterisks (*** p ≤ 0.001).

Figure 12

Identification rates for mean data of HI and NH listeners in Experiment 5 are plotted in the left and right panels, respectively. Probability of responding /da/ as a function of F₂-onset frequency is plotted separately for each absolute spectral tilt. Circles, asterisks, squares, x’s, upward triangles, downward triangles, and +’s represent mean data for CVs with absolute tilts of −12, −9, −6, −3, 0, +3, and +6 dB/oct. respectively. Maximum-likelihood fits of identification rates are displayed for mean data at each vowel tilt as different lines (see legend).

Figure 13

Standardized linear regression weights for tilt at each F₂ frequency for HI and NH listeners (filled squares and open circles, respectively) in Experiment 5. Error bars represent standard error of the estimate. Bonferroni-adjusted levels of statistical significance between groups at each frequency are indicated by asterisks (*** p ≤ 0.001).

Figure 14

Analysis of spectral tilt for sample stimuli from Experiments 1 – 4 (columns). Top row corresponds to input stimuli, middle row to recordings from a hearing aid with more or less uniform compression ratios set for a flat hearing loss configuration, and bottom row to recordings from a hearing aid with non-uniform compression ratios set for a sloping hearing loss configuration.