• ETRI Journal
• /
• v.32 no.1
• /
• pp.160-162
• /
• 2010

For the acoustic models of embedded speech recognition systems, hidden Markov models (HMMs) are usually quantized and the original full space distributions are represented by combinations of a few quantized distribution prototypes. We propose a maximum likelihood objective function to train the quantized distribution prototypes. The experimental results show that the new training algorithm and the link structure adaptation scheme for the quantized HMMs reduce the word recognition error rate by 20.0%.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.31B no.6
• /
• pp.17-24
• /
• 1994

This paper is a study on Korean word recognition and suggest the method that stabilizes the state-transition in the HMM by applying the `inertia' to the feature vector sequences. In order to reduce the quantized distortion considering probability distribution of input vectors, we used SOFM, an unsupervised learning method, as a vector quantizer, By applying inertia to the feature vector sequences, the overlapping of probability distributions for the response path of each word on the self organizing feature map can be reduced and the state-transition in the Hmm can be Stabilized. In order to evaluate the performance of the method, we carried out experiments for 50 DDD area names. The results showed that applying inertia to the feature vector sequence improved the recognition rate by 7.4% and can make more HMMs available without reducing the recognition rate for the SOFM having the fixed number of neuron.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.7 no.4
• /
• pp.649-655
• /
• 2006

This paper proposes a HMM-based recognition method using DMSVQ(Dynamic Multi-Section Vector Quantization) codebook by DMS model and double spectral feature, as a method on the speech recognition of speaker-independent. LPC cepstrum parameter is used as a instantaneous spectral feature and LPC cepstrum's regression coefficient is used as a dynamic spectral feature These two spectral features are quantized as each VQ codebook. HMM using DMS model is modeled by receiving instantaneous spectral feature and dynamic spectral feature by input. Other experiments to compare with the results of recognition experiments using proposed method are implemented by the various conventional recognition methods under the equivalent environment of data and conditions. Through the experiment results, it is proved that the proposed method in this paper is superior to the conventional recognition methods.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.31B no.9
• /
• pp.85-100
• /
• 1994

This paper proposes a recognition system of constrained handwritten Hangul and alphanumeric characters using discrete hidden Markov models (HMM). HMM process encodes the distortion and similarity among patterns of a class through a doubly stochastic approach. Characterizing the statistical properties of characters using selected features, a recognition system can be implemented by absorbing possible variations in the form. Hangul shapes are classified into six types by fuzzy inference, and their recognition is performed based on quantized features by optimally ordering features according to their effectiveness in each class. The constrained alphanumerics recognition is also performed using the same features used in Hangul recognition. The forward-backward, Viterbi, and Baum-Welch reestimation algorithms are used for training and recognition of handwritten Hangul and alphanumeric characters. Simulation result shows that the proposed method recognizes handwritten Korean characters and alphanumerics effectively.

• The Journal of the Acoustical Society of Korea
• /
• v.10 no.1
• /
• pp.52-60
• /
• 1991

This paper describes the study on isolated word recognition by using DHMM(Dynamic Hidden Markov Model) which has dynamic feature of spectrum as a parameter. This paper discusses speech recognition experiment basedon HMM which can evaluate not only instantaneous spectral features but also dynamic spectral features. LPC cepstrum parameters is used as a static feature and LPC cepstrum's regression coefficient is used as a dynamic feature. These two features are quantized by each VQ codebook. DHMM is modeled by receiving static vector and dynamic vector by input. In the whole experiment, as recognition experiment using DHMM shows 92.7% of recognition rate while the experiment using conventional HMM shows 88.8% of recognition rate, DHMM proved to be a useful model.