Abstract:

In second language (L2) listening activities, reproducing the diversity of speech information, such as talker variability, which exists in real-life communicative situations, is important for building a robust phonological database. As information and communication technology (ICT) use has increased during the COVID-19 pandemic, artificial intelligence (AI)-based synthetic speech and other technologies were developed, allowing learners to access a variety of L2 speeches. However, whether the use of synthetic speech is effective for L2 listening, particularly for perceptual learning and building a phonological database, remains unclear. This study examined the effectiveness of synthetic speech in L2 listening using an experiment on the mere exposure effect, which is closely related to perceptual learning. The mere exposure effect refers to favorability, or preference, increasing as the number of exposures increases. For instance, in advertising, this effect can be used to increase product preference by improving its visibility and exposure to potential customers. A well-known explanatory theory regarding this effect suggests that repeated exposure and increased perceptual fluency, or perceptual learning, are misattributed to favorability. The current experiment presented 25 Japanese university students, who were learners of English, with eight types of synthetic speech sounds: American and British English, with two male and two female voices each. Each learner listened to a set of 48 formulaic sequences with high familiarity, which is a subjective scale measuring the frequency of encountering, and 48 with low familiarity, with synthetic speech sounds changed in 24 sequences in each set. Each set was shown twice. The experiment was conducted in a computer room using PsychoPy®. Favorability was measured on a seven-point scale (1 = strongly dislike; 7 = strongly like). The results indicated that favorability was significantly higher for high-familiarity formulaic sequences when there was no change in voice ([BF = 4.361, error = 0 %] or, t[24] = 2.75, p = .011). These findings suggest that the use of synthetic speech may increase perceptual fluency and contribute to the acquisition of phonetic knowledge if the formulaic sequences are sufficiently familiar to learners. To increase the perceptual fluency of formulaic sequences with low familiarity, exposure frequency should be increased.

Keywords:

Synthetic speech, talker variability, mere exposure effect, favorability, perceptual learning.