• The Journal of the Acoustical Society of Korea
• /
• v.21 no.6
• /
• pp.559-565
• /
• 2002

In this paper, we proposed an efficient method that estimates the HMM (Hidden Marke Model) parameters of the noisy speech. In previous methods, noisy speech HMM parameters are usually obtained by analytical methods using the assumed noise statistics. However, as they assume some simplication in the methods, it is difficult to come closely to the real statistics for the noisy speech. Instead of using the simplication, we used some useful statistics from the clean speech HMMs and employed the deterministic noise model. We could find that the new scheme showed improved results with reduced computation cost.

• Speech Sciences
• /
• v.11 no.4
• /
• pp.75-88
• /
• 2004

In this paper, we propose a new approach to sequential linear regression adaptation of continuous density hidden Markov models (CDHMMs) based on transformation space model (TSM). The proposed TSM which characterizes the a priori knowledge of the training speakers associated with maximum likelihood linear regression (MLLR) matrix parameters is effectively described in terms of the latent variable models. The TSM provides various sources of information such as the correlation information, the prior distribution, and the prior knowledge of the regression parameters that are very useful for rapid adaptation. The quasi-Bayes (QB) estimation algorithm is formulated to incrementally update the hyperparameters of the TSM and regression matrices simultaneously. Experimental results showed that the proposed TSM approach is better than that of the conventional quasi-Bayes linear regression (QBLR) algorithm for a small amount of adaptation data.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.34S no.4
• /
• pp.57-65
• /
• 1997

The purpose of this paper is to propose the HMM (hidden markov model) based on multi-observation sequence for the isolated word recognition. The proosed model generates the codebook of MSVQ by dividing each word into several sections followed by dividing training data into several sections. Then, we are to obtain the sequential value of multi-observation per each section by weighting the vectors of distance form lower values to higher ones. Thereafter, this the sequential with high probability value while in recognition. 146 DDD area names are selected as the vocabularies for the target recognition, and 10LPC cepstrum coefficients are used as the feature parameters. Besides the speech recognition experiments by way of the proposed model, for the comparison with it, the experiments by DP, MSVQ, and genral HMM are made with the same data under the same condition. The experiment results have shown that HMM based on multi-observation sequence proposed in this paper is proved superior to any other methods such as the ones using DP, MSVQ and general HMM models in recognition rate and time.

• The Journal of the Acoustical Society of Korea
• /
• v.15 no.4E
• /
• pp.21-27
• /
• 1996

In this paper, we propose a discriminative training algorithm for the stochastic segment model (SSM) in continuous speech recognition. As the SSM is usually trained by maximum likelihood estimation (MLE), a discriminative training algorithm is required to improve the recognition performance. Since the SSM does not assume the conditional independence of observation sequence as is done in hidden Markov models (HMMs), the search space for decoding an unknown input utterance is increased considerably. To reduce the computational complexity and starch space amount in an iterative training algorithm for discriminative SSMs, a hybrid architecture of SSMs and HMMs is programming using HMMs. Given the segment boundaries, the parameters of the SSM are discriminatively trained by the minimum error classification criterion based on a generalized probabilistic descent (GPD) method. With the discriminative training of the SSM, the word error rate is reduced by 17% compared with the MLE-trained SSM in speaker-independent continuous speech recognition.

• Speech Sciences
• /
• v.8 no.3
• /
• pp.143-148
• /
• 2001

We propose a new scheme for VQ codebook search. The procedure is in between the binary-tree-search and full-search and thus might be called N-ary search of a codebook. Through the experiment performed on 7200 frames spoken by 25 speakers, we confirmed that the best codewords as good as by the full-search were obtained at moderate time consumption comparable to the binary-tree-search. In application to speech recognition by HMM/VQ with Bakis model, where appearance of a specific codeword is essential in the parameter training phase, the method proposed here is expected to provide an efficient training procedure.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.9 no.5
• /
• pp.543-548
• /
• 2014

Multiple-model based speech recognition framework (MMSR) has been known to be very successful in speech recognition. Since it uses multiple hidden Markov modes (HMMs) that corresponds to various noise types and signal-to-noise ratio (SNR) values, the selected acoustic model can have a close match with the test noisy speech. However, since the number of HMM sets is limited in practical use, the acoustic mismatch still remains as a problem. In this study, we experimentally determined the optimal SNR mapping between the test noisy speech and the HMM set to mitigate the mismatch between them. Improved performance was obtained by employing the SNR mapping instead of using the estimated SNR from the test noisy speech. When we applied the proposed method to the MMSR, the experimental results on the Aurora 2 database show that the relative word error rate reduction of 6.3% and 9.4% was achieved compared to a conventional MMSR and multi-condition training (MTR), respectively.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.5
• /
• pp.488-495
• /
• 2021

We propose a speech recognition system based on conformer. Conformer is known to be convolution-augmented transformer, which combines transfer model for capturing global information with Convolution Neural Network (CNN) for exploiting local feature effectively. The baseline system is developed to be a transfer-based speech recognition using Long Short-Term Memory (LSTM)-based language model. The proposed system is a system which uses conformer instead of transformer with transformer-based language model. When Electronics and Telecommunications Research Institute (ETRI) speech corpus in AI-Hub is used for our evaluation, the proposed system yields 5.7 % of Character Error Rate (CER) while the baseline system results in 11.8 % of CER. Even though speech corpus is extended into other domain of AI-hub such as NHNdiguest speech corpus, the proposed system makes a robust performance for two domains. Throughout those experiments, we can prove a validation of the proposed system.

• Proceedings of the KSPS conference
• /
• 2005.04a
• /
• pp.45-48
• /
• 2005

This study is about constructing L2 pronunciation correction system for L1 speakers using speech technology. Chinese pronunciation system consists of initials, finals and tones. Initials/finals are in segmental level and tones are in suprasegmental level. So different method could be used assessing Korean users' Chinese. The recognition rate of initials is 81.9% and that of finals is 68.7% in the standard acoustic model. Differ from native speech recognition, nonnative speech recognition could be promoted by additional modeling using L2 speakers' speech. As a first step for the those task we analysed nonnative speech and then set a strategy for modeling Korean speakers'.

• MALSORI
• /
• no.61
• /
• pp.101-111
• /
• 2007

A modified speech recogniser have been proposed for automatic capitalisation generation to improve the readability of English speech recognition output. In this modified speech recogniser, every word in its vocabulary is duplicated: once in a de-caplitalised form and again in the capitalised forms. In addition its language model is re-trained on mixed case texts. In order to evaluate the performance of the proposed system, experiments of automatic capitalisation generation were performed for 3 hours of Broadcast News(BN) test data using the modified HTK BN transcription system. The proposed system produced an F-measure of 0.7317 for automatic capitalisation generation with an SER of 48.55, a precision of 0.7736 and a recall of 0.6942.

• Journal of Digital Convergence
• /
• v.10 no.11
• /
• pp.377-382
• /
• 2012

In this paper, the echo noise robust CHMM learning model using echo cancellation average estimator LMS algorithm is proposed. To be able to adapt to the changing echo noise. For improving the performance of a continuous speech recognition, CHMM models were constructed using echo noise cancellation average estimator LMS algorithm. As a results, SNR of speech obtained by removing Changing environment noise is improved as average 1.93dB, recognition rate improved as 2.1%.