Vowel perception in noise

A comparison of children with cochlear implants and

their peers with typical hearing

Marcel Schlechtweg, Mark Gibson, Judit Ayala,

Xianhui Wang & Li Xu
Noise and spoken language
Table 1. Noise and the perception of spoken language: Overview of relevant factors
Hearing impairment and spoken language

➢ Surgically implanted electronic device

➢ Sound converted to electrical signal, directed to the

auditory nerve, bypassing damaged areas

➢ Severe-to-profound or moderate hearing loss / if

hearing aids are not efficient

➢ Sound impression different from typical hearing

(see, e.g., Battmer 2009; Clark 2003; Dang & Sweeney

2020; Dwyer & Gifford 2020; Faes & Gillis 2017; Gilbert
Figure 1. Ear with a cochlear implant & Pisoni 2012)
Hearing impairment and spoken language
➢ Despite benefits, several challenges remain

➢ Syntax (see, e.g., WaldmanDeLuca 2015)

➢ Morphology (see, e.g., Davies et al. 2020)
➢ Phonetics and Phonology (see, e.g., Basirat 2017)
➢ …
Focus on vowels I: Noise
➢ Confusions of articulatorily and acoustically similar vowels
in noise (see, e.g., Bent et al. 2010)

Figure 2. Vowel space (taken from https://www.internationalphoneticalphabet.org/html-ipa-keyboard-v1/keyboard/)

Focus on vowels II: Hearing impairment
➢ Vowel duration, F1, F2, and F3 can be affected
(production) (see, e.g., Hocevar-Boltezar et al. 2008;
Jafari et al. 2016; Kishon-Rabin et al. 1999; Nicolaidis &
Sfakianaki 2016; Nunez-Batalla et al. 2019; Palethorpe et
al. 2003; Rahilly 2012; Schenk et al. 2003a, 2003b)
Focus on vowels III: Noise and hearing
impairment
➢ Much less is known about vowel perception in noise (but
see Ferguson & Kewley-Port 2002; Iverson et al. 2006;
Munson & Nelson 2005; Winn 2011), especially for
children
Our experiment
Research question

How do children with two cochlear implants (CIs), in

comparison to their peers with typical hearing (TH),
perceive the five Spanish vowels [a], [e], [i], [o], and [u] in
different types of adverse listening conditions?
Participants
➢ 7 children with two CIs and 7 children with TH

➢ Matched for hearing age and biological sex

➢ Monolingual native speakers of Spanish

Table 2. Further characteristics of the children

Materials
➢ Five syllables [da], [de], [di], [do], [du]

➢ Spoken by one female and one male Spanish-speaking adult

➢ Embedded in two types of noise

➢ Multi-speaker background babble (six speakers)
(informational masker)
➢ Speech-shaped noise (energetic masker)

➢ Three SNRs (0, 6, 12)

Procedure
➢ Matlab-based test

➢ Five syllables presented on a computer screen

➢ Selection of syllable children heard

➢ 240 trials (5 vowels x 2 noise types x 3 SNRs x 2 speakers x

4 presentations each)

➢ Random distribution
Analysis I
➢ Descriptive analysis plus binomial logistic regression in R
(R Core Team 2021), using the packages lme4 and lmer
Test (Bates et al. 2015; Kuznetsova et al. 2017)

➢ Tukey tests (Lenth 2020)

➢ Response variable = Accuracy (in %)

Analysis I
➢ Fixed effects
➢ Group (children with CIs, children with TH)
➢ Vowel ([a], [e], [i], [o], [u])
➢ NoiseType (babble, speech-shaped)
➢ SNR (0, 6, 12)

➢ Random effect = Intercept by Participant

➢ First model = Only main effects

➢ Second model = With all possible two-way interactions

Analysis I

➢ Non-significant terms removed from the model

➢ In case of convergence issues: Change of optimizer

Analysis II
➢ Error type analysis (types of vowel confusions)
Results
Figure 3. Main Figure 4. Main
effect of Group effect of Vowel

All comparisons significant

except for [a] vs. [e]

Figure 5. Main Figure 6. Main

effect of effect of SNR
NoiseType

All comparisons significant

Results
Significant interaction

➢ With the exception of [a] and [u], children with CIs

responded significantly less accurately than
children with TH

➢ Children with CIs only: Significantly higher accuracy

for [i] than for [o]

➢ Children with TH only: Significantly higher accuracy

for [o] than for [u]

Figure 7. Interaction of Group and Vowel

Results
Significant interaction

➢ Significantly higher accuracy for speech-shaped

than for babble noise at 0 dB only

Figure 8. Interaction of NoiseType and SNR

Results
Table 3. Error types
Summary and discussion
➢ Main effects for Group and SNR are in line with many
previous findings (e.g., Fowler et al. 2021; Jin & Liu 2014)

➢ Main effect of NoiseType (and the direction) finds some

support (see Garcia Lecumberri & Cooke 2006; Regalado et
al. 2019; Ren et al. 2018)

➢ Main effect of Vowel partly compatible with findings based on

listeners with TH (see Cutler et al. 2004)
Summary and discussion
➢ Lack of visual support possibly especially problematic for the
perception of rounded (back) vowels (see Robert-Ribes et al.
1998)

➢ Problem with [o] and [u] also reported in the literature (see
Välimaa et al. 2002)

➢ Order of Spanish vowel acquisition reflected in our data (see

Hernández Pina 1984 cited in Lléo 2012)
Summary and discussion
➢ Children with CIs seem to have difficulties in perceiving
vowels on more acoustic levels (F1, F2) in noise than their
peers with TH