A Word Shadowing Error Detection Method Based on Kalman Filtering

Word shadowing error detection has wide applications in language learning and speech assessment. However, due to variations in speech rate, individual pronunciation differences, and dynamic feature misalignment, traditional speech assessment methods struggle to achieve high-precision real-time error correction. This paper proposes a new word shadowing error detection method based on Kalman filtering to enhance the robustness and accuracy of pronunciation error detection. The basic idea is that the difference between reference waveform and test waveform should maintain a stable and smooth curve. The Kalman filter can adaptively detect mismatches, especially in situations where individual pronunciation differences are significant. To this end, Two speeches are firstly aligned using the Dynamic Time Warping (DTW) algorithm. Then a differential operation is used to depict the voice trend along time. Finally, based on the voice trend, a detector via Kalman filtering is used to predict the difference variance to detect error points. Experimental results indicate that the proposed method effectively improves the accuracy of word shadowing error detection while significantly reducing false detection rates.