On the beneficial influence of perceptual knowledge in pronunciation error detection
We discuss a general approach to assess the pronunciation of individual foreign speakers using a psychoacoustic model. The use of auditory information can contribute significantly towards a better distinction of the phonetically relevant speech elements. Listeners exhibit a great ability to distinguishing between native and non-native speech based entirely on its characteristics. The human hearing system has the capacity to perceive a wide range of sound distinctions, including those that occur in a language. This research investigates which aspects of the speech signal are captured by a native listener. Studying a speech signal representation does not necessarily lead to finding the potential mispronunciation. Let’s for argument’s sake assume that two native speakers produce the same word. Likely, their speech signals are not identical, though they both considered as being natively pronounced. It is their perceptual characteristics that classify the above two utterances into the native speech category. By establishing the native-perceptual assessment degree (NatePAD) we find the mispronounced phonemes by non-native speakers.