The role of foreign accent and short-term exposure in speech-in-speech recognition

Abstract

Daily speech communication often takes place in suboptimal listening conditions, in which interlocutors typically need to segregate the target signal from the background sounds. The present study investigated the influence on speech recognition of a relatively familiar foreign accent in background speech (Exp. 1) and whether short-term immediate exposure to the target talker’s voice (Exp. 2) or the background babble (Exp. 3) would either help or hinder the segregation of target from background. A total of 72 native Dutch participants were asked to listen to Dutch target sentences in the presence of Dutch or German-accented Dutch babble without (Exp. 1) or with (Exps. 2 and 3) an exposure phase. Their task was to write down what they heard. The results of Experiment 1 revealed that listeners gained a release from masking when the background speech was accented, indicating that dissimilar and less familiar signals are easier to segregate effectively. Experiment 2 demonstrated that short-term immediate exposure to the target talker had no effect on speech-in-speech recognition, whereas exposure to the background babble could hinder separating the target voice from the background speech (Exp. 3). However, this reduced release from masking only appeared in the more difficult and more familiar babble condition (Dutch in Dutch), in which the speech recognition system may have remained attuned to the babble as a potential source of communicatively relevant information. Overall, this research provides evidence that both short-term adaptation and the degrees of target–background similarity and familiarity are of importance for speech-in-speech recognition.