Assessment of Non-native Prosody for Spanish as L2 using quantitative scores and perceptual evaluation

Abstract

In this work we present SAMPLE, a new pronunciation database of Spanish as L2, and first results on the automatic assessment of Non-native prosody. Listen and repeat and read tasks are carried out by native and foreign speakers of Spanish. The corpus has been designed to support comparative studies and evaluation of automatic pronunciation error assessment both at phonetic and prosodic level. Four expert evaluators have annotated utterances with perceptual scores related to prosodic aspects of speech, intelligibility, phonetic quality and global proficiency level in Spanish. From each utterance, we computed several prosodic features and ASR scores. A correlation study over subjective and quantitative measures is carried out. An estimation of the prediction of perceptual scores from speech features is shown.

Resources

Details