Abstr act. This paper aims to design and optimize the algorithm of Chinese isolated words speech recognition, to improve the recognition accuracy and robustness. Mel-Frequency Cepstral Coefficient (MFCC) has been widely used in the art-of-state speech recognition since it considers the characteristics of human voice and sounds receiving, and has better robustness for the recognition accuracy. The MFCCs of different speech frame are relevant, and each Chinese isolated word has its own optimal number of codebook. In this paper, we use the first-order differential and second-order differential of MFCC to improve the recognition accuracy and robustness, and the results show that the difference between the first recognition probability and the second recognition probability increases. By establishing an optimal codebook for each isolated word, and using codebook adaptive algorithm, the robustness of recognition is also improved greatly.
[1]
Antonio M. Peinado.
Speech Recognition Over Digital Channels: Robustness and Standards
,
2006
.
[2]
Tuan D. Pham,et al.
Fuzzy declustering-based vector quantization
,
2009,
Pattern Recognit..
[3]
D. Burton,et al.
Isolated-word speech recognition using multisection vector quantization codebooks
,
1984,
IEEE Trans. Acoust. Speech Signal Process..
[4]
Fatma Zohra Chelali,et al.
MFCC and vector quantization for Arabic fricatives speech/speaker recognition
,
2012,
2012 International Conference on Multimedia Computing and Systems.
[5]
Wendy J. Holmes,et al.
Speech Synthesis and Recognition
,
1988
.
[6]
Richard J. Mammone,et al.
New LP-derived features for speaker identification
,
1994,
IEEE Trans. Speech Audio Process..