High variability phonetic training in adaptive adverse conditions is rapid, effective, and sustained

Abstract

This paper evaluates a novel high variability phonetic training paradigm that involves presenting spoken words in adverse conditions. The effectiveness, generalizability, and longevity of this high variability phonetic training in adverse conditions was evaluated using English phoneme contrasts in three experiments with Malaysian multilinguals. Adverse conditions were created by presenting spoken words against background multi-talker babble. In Experiment 1, the adverse condition level was set at a fixed level throughout the training and in Experiment 2 the adverse condition level was determined for each participant before training using an adaptive staircase procedure. To explore the effectiveness and sustainability of the training, phonemic discrimination ability was assessed before and immediately after training (Experiments 1 and 2) and 6 months after training (Experiment 3). Generalization of training was evaluated within and across phonemic contrasts using trained and untrained stimuli. Results revealed significant perceptual improvements after just three 20-minute training sessions and these improvements were maintained after 6 months. The training benefits also generalized from trained to untrained stimuli. Crucially, perceptual improvements were significantly larger when the adverse conditions were adapted before each training session than when it was set at a fixed level. As the training improvements observed here are markedly larger than those reported in the literature, this indicates that the individualized phonetic training regime in adaptive adverse conditions (HVPT-AAC) is highly effective at improving speech perception.

Cross-talker generalization in the perception of nonnative speech: A large-scale replication.

“…Research on generalization in speech perception, language learning, and beyond has put a strong emphasis on the role of “variability” during exposure in determining the success of learning (Aguilar et al, 2018; Bradlow, 2008; Giannakopoulou et al, 2017; Leong et al, 2018; Sadakata & McQueen, 2014). One key insight of these works has been that increased variability during exposure often furthers, rather than hinders, learning and generalization.…”

Auditory category learning is robust across training regimes

“…For example, high variability phonetic training consisting of speech exemplars from multiple speakers and across multiple word forms can improve non-native category learning and generalization (Logan et al, 1991). More generally, greater acoustic variability of the exemplars can lead to learning improvements for speech category learning in people's first language (Estes & Lew-Williams, 2015;Galle et al, 2015;Rost & McMurray, 2009Singh, 2008) and second language (Barcroft & Sommers, 2005;Bradlow et al, 1999;Leong et al, 2018;Lim & Holt, 2011;Lively et al, 1993;Shinohara & Iverson, 2018). Past studies have employed various techniques to introduce high acoustic variability, such as using multiple talkers, a single talker with high acoustic variability, multiple prosodic voice affectations, multiple word forms, and variability across a category-diagnostic acoustic dimension.…”

Xie, Liu, & Jaeger (2020). Cross-talker generalization during foreign-accented speech perception

“…Research on generalization in speech perception, language learning and beyond has put a strong emphasis on the role of "variability" during exposure in determining the success of learning (Aguilar, Plante & Sandoval, 2018;Bradlow, 2008;Giannakopoulou, Brown, Clayards & Wonnacott, 2017;Leong, Price, Pitchford & van Heuven, 2018;Sadakata & Mc-Queen, 2014). One key insight of these works has been that increased variability during exposure often furthers, rather than hinders, learning and generalization.…”