Detection of Lexical Stress Errors in Non-Native (L2) English with Data Augmentation and Attention
This work is significant for L2 English learners, particularly Slavic and Baltic speakers, by improving the automatic detection of lexical stress errors, which can aid in pronunciation training.
This paper addresses the detection of lexical stress errors in non-native English speech by introducing an attention-based deep learning model for syllable-level feature extraction and a data augmentation technique using Neural Text-To-Speech (TTS) to generate incorrectly stressed words. The combined approach achieved 94.8% precision and 49.2% recall for detecting incorrectly stressed words in L2 English speech from Slavic and Baltic speakers.
This paper describes two novel complementary techniques that improve the detection of lexical stress errors in non-native (L2) English speech: attention-based feature extraction and data augmentation based on Neural Text-To-Speech (TTS). In a classical approach, audio features are usually extracted from fixed regions of speech such as the syllable nucleus. We propose an attention-based deep learning model that automatically derives optimal syllable-level representation from frame-level and phoneme-level audio features. Training this model is challenging because of the limited amount of incorrect stress patterns. To solve this problem, we propose to augment the training set with incorrectly stressed words generated with Neural TTS. Combining both techniques achieves 94.8% precision and 49.2% recall for the detection of incorrectly stressed words in L2 English speech of Slavic and Baltic speakers.