The effect of hyperarticulation on speech comprehension under adverse listening conditions

Abstract

Comprehension assesses a listener’s ability to construe the meaning of an acoustic signal in order to be able to answer questions about its contents, while intelligibility indicates the extent to which a listener can precisely retrieve the acoustic signal. Previous comprehension studies asking listeners for sentence-level information or narrative-level information used native listeners as participants. This is the first study to look at whether clear speech properties (e.g. expanded vowel space) produce a clear speech benefit at the word level for L2 learners for speech produced in naturalistic settings. This study explored whether hyperarticulated speech was more comprehensible than non-hyperarticulated speech for both L1 British English speakers and early and late L2 British English learners in quiet and in noise. Sixteen British English listeners, 16 native Mandarin Chinese listeners as early learners of L2 and 16 native Mandarin Chinese listeners as late learners of L2 rated hyperarticulated samples versus non-hyperarticulated samples in form of words for comprehension under four listening conditions of varying white noise level (quiet or SNR levels of + 16 dB, + 12 dB or + 8 dB) (3 × 2× 4 mixed design). Mean ratings showed all three groups found hyperarticulated speech samples easier to understand than non-hyperarticulated speech at all listening conditions. Results are discussed in terms of other findings (Uther et al., 2012) that suggest that hyperarticulation may generally improve speech processing for all language groups.

Clear speech in Tashlhiyt Berber: The perception of typologically uncommon word-initial contrasts by native and naive listeners

“…One possible explanation for the high performance of naive listeners in these conditions is that our stimuli were elicited to a non-native Tashlhiyt listener. Indeed, prior work has shown that non-native-directed speech is better perceived by both native and non-native listeners (Kangatharan et al, 2022). While the present study looked only at non-native directed speech, an avenue for future work is to investigate how clear speech patterns vary across different types of listeners and with different types of instructions.…”

The effect of clear speech to foreign-sounding interlocutors on native listeners’ perception of intelligibility

“…This study had two parts: a speech production experiment to elicit spontaneous speech produced when doing a 'Spot the Difference' task with different types of interlocutors as described previously in Kangatharan et al (2021), and a perceptual evaluation experiment using target words extracted from the speech production experiment. Both the speech production experiment and the perceptual evaluation experiment were approved by the Ethics committee at the Psychology Department at Brunel University.…”

“…This speech production experiment was conducted in a previous study (Kangatharan et al, 2021), and details are reported again here in the present study to describe how the speech tokens used for the speech perception experiment were obtained. The speech production experiment used a 2 (interlocutor's accent: native, foreign) × 2 (interlocutor's physical appearance: native, foreign) × 3 (three target vowels: /a:/, /uː/ and /iː/) mixed design with four different types of interlocutors: NLNS (native-looking and native-sounding), NLFS (native-looking and foreign-sounding), FLNS (foreign-looking and native-sounding), and FLFS (foreign-looking and foreign-sounding).…”

“…The speech production experiment used a 2 (interlocutor's accent: native, foreign) × 2 (interlocutor's physical appearance: native, foreign) × 3 (three target vowels: /a:/, /uː/ and /iː/) mixed design with four different types of interlocutors: NLNS (native-looking and native-sounding), NLFS (native-looking and foreign-sounding), FLNS (foreign-looking and native-sounding), and FLFS (foreign-looking and foreign-sounding). The extent of hyperarticulation in the target words in which one of the three target vowels was present was the dependent variable (see Kangatharan et al, 2021). In the speech production study, fifty-two female White British speakers were instructed to interact with an interlocutor from one of the four different interlocutor types.…”

“…The present study used three different measures of intelligibility with naturally elicited speech samples collected from a previous speech production experiment in which speech directed to foreign-sounding interlocutors and speech directed to native-sounding interlocutors were recorded (Kangatharan et al, 2021). These measures of intelligibility consist of an orthographic transcription task (together with a confidence rating), a typicality rating task and a clarity rating task.…”