—Pronunciation scoring tries to evaluate on voice
segmentation objectively with related recognition technologies.
The voice to be evaluated (VBE) was compared with standard
voice (SV) to arrive at a score in accordance with similarity.
The system firstly extracted feature parameters from VBE and
SV. Two parameters of linear predictive cepstral coefficients
(LPCC) and Mel-frequency cepstral coefficients (MFCC) were
selected. The distance between SV vector and VBE vector was
computed with dynamic time warping (DTW) algorithm.
According to scoring formula, the obtained distance was
translated into score, which can be regarded as evaluation on
VBE. Experiment shows that expert scoring and system scoring
have a high correlation by combining LPCC and MFCC.
—Pronunciation scoring, feature parameter,
dynamic time warping.
Jie Yang is with the Shanghai Second Polytechnic University, China
Cite:Jie Yang, "Toward a Pronunciation Scoring Method Based on Multi-Feature Fusion," International Journal of Computer Theory and Engineering vol. 7, no. 4, pp. 264-267, 2015.