Words Get in the Way: Linguistic Effects on Talker Discrimination

Abstract

A speech perception experiment provides evidence that the linguistic relationship between words affects the discrimination of their talkers. Listeners discriminated two talkers' voices with various linguistic relationships between their spoken words. Listeners were asked whether two words were spoken by the same person or not. Word pairs varied with respect to the linguistic relationship between the component words, forming either: phonological rhymes, lexical compounds, reversed compounds, or unrelated pairs. The degree of linguistic relationship between the words affected talker discrimination in a graded fashion, revealing biases listeners have regarding the nature of words and the talkers that speak them. These results indicate that listeners expect a talker's words to be linguistically related, and more generally, indexical processing is affected by linguistic information in a top-down fashion even when listeners are not told to attend to it.

How many voices did you hear? Natural variability disrupts identity perception from unfamiliar voices

“…Furthermore, exemplars differed in their linguistic/verbal content (different utterances within and across voice identity), verbal register, type of utterance, vocal effort (quiet conversation vs. shouting) as well as their perceived affective properties, such as valence and arousal, and perceived likeness, among any number of features. While the current study shows how uncontrolled natural within‐person variability from a range of sources can affect speaker identity perception, other studies have shown how specific sources of variability can affect perception (e.g., language spoken, Zarate et al ., ; linguistic content, Narayan et al ., ; vocalizations type, Lavan et al ., ; distinctiveness, Papcun et al ., ; and duration of the exemplars Schweinberger, Herholz, & Sommer, ). How these different types of variability relate to each other and interact in the context of identity perception is largely unexplored.…”

“…For example, listeners are less accurate at correctly matching speakers across pairs of sentences produced in different languages compared to when pairs include the same language (Wester, ; Zarate, Tian, Woods, & Poeppel, ). Furthermore, linguistic (dis)similarity of stimuli affects speaker discrimination performance in a top‐down fashion: Identities can be more accurately discriminated from pairs of stimuli that are semantically or phonetically related, such as ‘day‐dream’ or ‘day‐bay’, than from linguistically unrelated stimuli, such as ‘day‐bee’ (Narayan, Mak, & Bialystok, ). Similarly, listeners fail to reliably discriminate between unfamiliar identities when making judgements for pairs of disguised and undisguised voices (e.g., hypernasal voice vs. neutral voice; Reich & Duke, ), across different vocalizations (e.g., vowels vs. laughter; Lavan et al ., ), and across sung versus spoken words (Peynircioğlu, Rabinovitz, & Repice, ).…”

Forensic voice discrimination by lay listeners: The effect of speech type and background noise on performance

“…Task difficulty may increase the likelihood that lay listeners will attribute the variability to the same identity because they lack expert knowledge about how individual voices can vary across instances and are unable to isolate high‐level features of speech that are stable across utterances (Alexander et al, ; Leemann et al, ). From a perceptual point of view, the presence of intraspeaker variability is problematic for lay listeners trying to discriminate between identities (Lavan et al, ; Narayan et al, ; Wester, ). This is supported by the pattern of results observed in these experiments.…”

“…However, what is clear from the previous literature is that stimulus variability negatively affects voice discrimination performance. Participants are less accurate when making judgments across different vocalizations (Lavan et al, ), different languages (Wester, ), and when linguistic content varies across to‐be‐compared samples (Narayan, Mak, & Bialystok, ).…”

Breaking voice identity perception: Expressive voices are more confusable for listeners

“…Finally, speaker identification and discrimination are more challenging in speech in an unfamiliar language compared to a familiar language (Zarate, Tian, Woods & Poeppel, 2015;Winters, Levi & Pisoni, 2008). Unfamiliar listeners have furthermore been shown to struggle to accurately generalise identity information across signals that include within-person variability: performance drops when listeners made judgements across different languages (Wester, 2012), divergent linguistic content (Narayan, Mak & Bialystok, 2017), different types of vocalisation (Lavan et al, 2016), disguised and undisguised voices (Reich & Duke, 1979) and across sung vs. spoken speech (Peynircioğlu, Rabinovitz, & Repice, 2017). While voice identity processing for familiar voices is usually more robust to disruption introduced by within-person variability, there are nonetheless striking examples of when familiar voice processing fails: familiar individuals are not well recognised when speaking in a falsetto voice (Wagner & Köster, 1999) or when listeners make speaker discrimination judgements across different types of vocalisation (Lavan et al, 2016).…”