The Temporal Fine Structure of Background Noise Determines the Benefit of Bimodal Hearing for Recognizing Speech

Abstract

Cochlear implant (CI) users have more difficulty understanding speech in temporally modulated noise than in steady-state (SS) noise. This is thought to be caused by the limited low-frequency information that CIs provide, as well as by the envelope coding in CIs that discards the temporal fine structure (TFS). Contralateral amplification with a hearing aid, referred to as bimodal hearing, can potentially provide CI users with TFS cues to complement the envelope cues provided by the CI signal. In this study, we investigated whether the use of a CI alone provides access to only envelope cues and whether acoustic amplification can provide additional access to TFS cues. To this end, we evaluated speech recognition in bimodal listeners, using SS noise and two amplitude-modulated noise types, namely babble noise and amplitude-modulated steady-state (AMSS) noise. We hypothesized that speech recognition in noise depends on the envelope of the noise, but not on its TFS when listening with a CI. Secondly, we hypothesized that the amount of benefit gained by the addition of a contralateral hearing aid depends on both the envelope and TFS of the noise. The two amplitude-modulated noise types decreased speech recognition more effectively than SS noise. Against expectations, however, we found that babble noise decreased speech recognition more effectively than AMSS noise in the CI-only condition. Therefore, we rejected our hypothesis that TFS is not available to CI users. In line with expectations, we found that the bimodal benefit was highest in babble noise. However, there was no significant difference between the bimodal benefit obtained in SS and AMSS noise. Our results suggest that a CI alone can provide TFS cues and that bimodal benefits in noise depend on TFS, but not on the envelope of the noise.

Keywords: bimodal hearing; cochlear implants; hearing aids; sensorineural hearing loss; speech intelligibility; speech perception.

Fig. 1

The three noise types used for speech-in-noise testing: steady-state (SS) noise (a), babble noise (b), and amplitude-modulated steady-state (AMSS) noise (c). AMSS was generated by modulating SS noise with the envelope of the babble noise (red lines). Insets show details of babble and AMSS noise

Fig. 2

Median pure-tone audiogram for the non-implanted ear of all subjects. Shaded areas indicate the interquartile range. Hatched box: exclusion criterion based on residual hearing

Fig. 3

Frequency spectra determined by fast Fourier transformation of the speech material (green), steady-state (SS) noise (blue), babble noise (red), amplitude-modulated SS (AMSS) noise (purple), and the babble noise envelope (black)

Fig. 4

Test environment. Schematic of the homogeneous noise setup (a) and frequency characteristics of the loudspeakers (b). Speech was presented in front through a center loudspeaker and noise was presented through 8 loudspeakers positioned around the listener. The frequency characteristics shown in (b) were recorded in 1/3 octave bands of the surround loudspeakers (solid red line) used to present noise, and the center loudspeaker (dashed blue line) used for presenting the target speech stimuli. The stimulus was pink noise that was calibrated at the same overall sound level for the center and surround loudspeakers. Vertical lines: low-pass (350 Hz) and high-pass cutoff (8.7 kHz) of the CI speech processor (horizontal arrow labeled “CI”). The HA had an acoustic range (horizontal arrow labeled “HA”) with a lower low-pass cutoff (125 Hz), and a similar high-pass cutoff (Advanced Bionics LLC 2016)

Fig. 5

Effects of the 8-talker babble on modulation depth. 10-s fragments of the original ICRA babble noise (a) and the corresponding 8-talker babble noise as recorded from a CI and a KEMAR manikin (b) showing a reduction in the depth of amplitude modulation in the 8-talker condition

Fig. 6

SRTs and bimodal benefits in 15 subjects. a SRTs were obtained in steady-state (SS) noise (green circles) and babble noise (purple circles) when listening with a cochlear implant only (CI), with a HA only (HA), and with both a CI and HA (CI + HA). Lower SRTs represent better speech recognition. b Corresponding bimodal benefits; higher values represent larger bimodal benefits. *P < 0.05

Fig. 7

SRTs and bimodal benefits in 11 subjects. a SRTs were obtained in steady-state (SS) noise, amplitude-modulated steady-state (AMSS) noise, and babble noise with a cochlear implant only (CI, red circles) and with both a CI and HA (CI + HA, blue circles). Lower SRTs represent better speech recognition. b Corresponding bimodal benefits obtained in SS (green circles), AMSS (brown circles), and babble noise (purple circles). Higher values represent larger bimodal benefits. *P < 0.05

Fig. 8

Difference SRT (SRT in babble noise (SRT_babble) minus SRT in AMSS noise (SRT_AMSS)) plotted against the average pure-tone threshold across the frequencies 125, 250, and 500 Hz (PTA_125–500) in the CI-only condition. The linear correlation was not significant (F(1,9) = 0.84, P = 0.38)

Fig. 9

Computer-simulation of the CI pulse-train output to babble noise and amplitude-modulated steady-state (AMSS) noise. A short fragment of acoustic babble noise (a) and the corresponding AMSS noise (b) and their corresponding pulse-train outputs are shown for electrodes 2 and 3 (c, d), 8 and 9 (e, f), and 14–15 (g, h). The fragment resembled the consonant [s], with a prominent high-frequency component in it. The envelope of the babble noise was used to amplitude-modulate the steady-state state noise, yielding AMSS noise with a more prominent intermediate-frequency component. Electrodes in Advanced Bionics arrays are numbered from apical (1) to basal (16). Because of the current steering used in the HiRes strategies from Advanced Bionics, pulse trains are generated by activating pairs of electrodes