Identifiability

The second main aim of this dissertation is to investigate the identifiability of HI children’s speech. The main research question is whether listeners can differentiate between the speech of NH and HI children. This section elaborates on previous studies on identifiability (§1.5.1) and on the relevance of studying the identifiability in HI children’s speech (§1.5.2). Moreover, it points out some difficulties in investigating identifiability through a labelling task (§1.5.3), leading to the concept of speech quality assessment (§1.5.4).

1.5.1 Markers of identity in speech

Studies on identifiability implicitly all have the same starting point:

speech signals do not only contain linguistic cues on the content of an utterance, but also provide the listener with contextual cues on the speaker. That is, on top of the actual message, listeners are able to hear physical, psychological and social markers of the individual speaker (Laver

& Trudgill, 1979).

For example, speech reveals biological features such as the gender, age and overall physique of the speaker. Since there is a clear correlation between these features and an individual’s vocal apparatus, listeners can reliably predict these features (Laver & Trudgill, 1979). However, less obvious individual characteristics are also hidden in the speech stream and potentially detectable for the listener. For example, non-native accents are reliably distinguished from native accents (Doeleman, 1998; Flege, 1984) and within one language, listeners are able to differentiate between local accents and thus, categorising speakers into groups (Grondelaers et al., 2011; Grondelaers et al., 2010). Objectively, the fact that listeners are capable of identifying these aspects implies that speech contains particular features that give away some characteristics of the speaker. Subjectively, listeners have stereotypical preconceptions of non-native accents in terms of the attractiveness and the status of the accent. For example, Dutch listeners find a Turkish, Moroccan or German accent in Dutch less attractive than for example an English accent (Doeleman, 1998; Russo et al., 2017). Similarly, regional (native) accents can be perceived as more or less attractive by other native speakers of that language. Since listeners

could also take this subjective perception into account when judging HI speech, the present dissertation includes native as well as non-native listeners in the speech identifiability experiments.

1.5.2 Markers of hearing status in hearing-impaired children

Studies on the speech of HI children have shown that children who receive their hearing device at a very young age seem to catch up on linguistic aspects. Also with respect to their speech production, some studies suggest that these children reach age-appropriate speech intelligibility (Bakhshaee et al., 2007; Chin et al., 2012; De Raeve, 2010; Toe

& Paatsch, 2013). But when these children enter primary school, is their speech distinguishable from the speech of normally hearing peers? In other words, is their speech still identifiable as that of a hearing-impaired individual? Previous studies have shown that the speech of children with CI and HA exhibits particular (acoustic) deviances from NH children’s speech, even in early rehabilitated children with several years of device use (Baudonck et al., 2011a; Baudonck et al., 2015; Lenden & Flipsen, 2007;

Peng et al., 2008; Van Lierde et al., 2005; Verhoeven et al., 2016). Possibly, these deviances are picked up by the listeners, which would be in line with Miller (2013: 601) who stated: “Listeners readily register when someone has speech that draws attention to itself, when pronunciation deviates from some notion of normal. Even when intelligibility is not particularly affected, altered sound production may be linked with marked psychosocial consequences for the speaker”. Until now, no other study has investigated the speech identifiability of HI children’s speech. In this

dissertation, speech identifiability will be investigated by means of a labelling identification task (§1.5.3) and overall speech quality judgements (§1.5.4).

1.5.3 Labelling identification task

Speaker identifiability is usually assessed by means of a labelling task. In this task, stimuli are presented one by one to the listener who is required to label each stimulus. These labels are preset categories that are communicated to the listener at the start of the experiment (McGuire, 2010).

For studies on identifiability, a labelling task is a simple and yet effective method. However, the approach also has some inherent difficulties. First of all, the listeners have to be informed about the possible categories and thus, the topic of the investigation is revealed. Moreover, knowing about the categories of interest possibly activates stereotypes in the listeners which they could transmit in their judgements (Doeleman, 1998). Secondly and maybe most importantly, a labelling task requires that the listeners know how a certain category corresponds to – in this case – a certain type of speech. In other words, a labelling task is very difficult if the listeners do not recognise the labels and what they represent. For example, if listeners have to identify the regional background of Belgian Dutch speakers, they need to at least have a mental representation of the varieties represented in the labels such as East Flemish and West Flemish (Grondelaers & Lybaert, 2017). Related to this point, if the listeners are uncertain about categories and labels, they might confuse categories. For

example, if listeners confuse East and West Flemish, they consistently mismatch the mental representation and the labels. Also, their judgements can be based on false presumptions or stereotypes.

In line with the studies on accent identifiability (Grondelaers &

Lybaert, 2017; Grondelaers et al., 2011), the differentiation of categories if labels are unknown to the listeners can entail difficulties when labelling (HI) children’s speech. In such a labelling task, the preset categories that are communicated to the listeners are the hearing statuses NH, CI and HA.

The latter two labels, i.e. CI and HA, share the characteristic “HI” and, thus, the differentiation can be perceived as vague for the listeners (Alfonso-Reese et al., 2002). This possibility should be considered in the labelling task.

1.5.4 Speech quality judgements

Speech quality judgements are a possible alternative to the traditional labelling task. Similar to speech intelligibility, speech quality can be examined from two perspectives. On the one hand, speech quality can refer to the quality of an incoming signal. Particularly in hearing aid users, aspects such as noise affect how they perceive the quality of speech (Kressner et al., 2013). On the other hand, speech quality is a measure of how listeners rate the naturalness or acceptability of an individual’s speech (Maassen & Povel, 1985). In other words, speech quality judgements refer to a listener’s general impression of the quality of speech (Kondo, 2012).

This dissertation focuses on the latter perspective.

Studies on the speech of children with HI have focused on both detailed aspects of their speech as well as more general outcome measures such as speech intelligibility. Compared to speech intelligibility, investigating the overall speech quality of HI children takes an even more general approach to these children’s speech. Until now, there are very few studies on HI children’s speech quality. In the 1980s, Maassen and Povel (1985) investigated the effect of (supra-)segmental manipulation on the speech intelligibility and speech quality of deaf children. In this study, both measures were highly correlated. Moreover, in a study on the vowel production of HI children, Baudonck et al. (2011c: 159) proposed to examine speech quality: “[T]he observed subtle acoustic formant distinctions, some of which can remain undetected by the naked ear, may perhaps signal deficits in ‘naturalness’ of a child’s speech. When it comes to vowel production in early-rehabilitated hearing-impaired children, further research should pay more attention to speech ‘naturalness’ rather than to intelligibility”. To the best of our knowledge, no other study suggested or examined the overall speech quality in CI and HA children.

The labelling task and the speech quality judgements both provide an answer to the question whether listeners perceive a difference between the speech of NH, CI and HA children. In the labelling task, listeners are asked to categorise speech utterances. Speech quality judgements enable a more indirect approach without labels. As a result, listeners do not need knowledge on the hearing statuses and they cannot consider presumptions or stereotypes in their judgements.