• The Journal of the Acoustical Society of Korea
• /
• v.17 no.2E
• /
• pp.31-37
• /
• 1998

This paper proposes a feature extraction method using wavelet transform for speech recognition. Speech recognition system generally carries out the recognition task based on speech features which are usually obtained via time-frequency representations such as Short-Time Fourier Transform (STFT) and Linear Predictive Coding(LPC). In some respects these methods may not be suitable for representing highly complex speech characteristics. They map the speech features with same may not frequency resolutions at all frequencies. Wavelet transform overcomes some of these limitations. Wavelet transform captures signal with fine time resolutions at high frequencies and fine frequency resolutions at low frequencies, which may present a significant advantage when analyzing highly localized speech events. Based on this motivation, this paper investigates the effectiveness of wavelet transform for feature extraction of wavelet transform for feature extraction focused on enhancing speech recognition. The proposed method is implemented using Sampled Continuous Wavelet Transform (SCWT) and its performance is tested on a speaker-independent isolated word recognizer that discerns 50 Korean words. In particular, the effect of mother wavelet employed and number of voices per octave on the performance of proposed method is investigated. Also the influence on the size of mother wavelet on the performance of proposed method is discussed. Throughout the experiments, the performance of proposed method is discussed. Throughout the experiments, the performance of proposed method is compared with the most prevalent conventional method, MFCC (Mel0frequency Cepstral Coefficient). The experiments show that the recognition performance of the proposed method is better than that of MFCC. But the improvement is marginal while, due to the dimensionality increase, the computational loads of proposed method is substantially greater than that of MFCC.

• Phonetics and Speech Sciences
• /
• v.10 no.1
• /
• pp.1-7
• /
• 2018

Categorical production in language processing helps speakers to produce phonemic contrasts. This categorization and production is utilized for the production-based and imitation-based approach in the present study. Contrastive signals in speakers' speech reflect the shapes of boundaries with categorical characteristics. Signals that provide information about lexical pitch accent contrasts can introduce categorical distinctions for productive and cognitive selection. This experiment was conducted with nine North Kyungsang speakers for a production task and nine North Kyungsang speakers for an imitation task. The first finding of the present study is the rigidity of categorical production, which controls the boundaries of lexical pitch accent contrasts. The categorization of North Kyungsang speakers' production allows them to classify minimal pitch accent contrasts. The categorical production in imitation appeared in two clusters, representing two meaningful contrasts. The second finding of the present study is that there are individual differences in speakers' production and imitation responses. The distinctive performances of individual speakers showed a variety of curves. For the HL-LH patterns, the categorical production tended to be highly distinctive as compared to the other pitch accent patterns (HH-HL and HH-LH), showing that there are more continuous curves than categorical curves. Finally, the present study shows that, for North Kyungsang speakers, imitative production is the core type of categorical production for determining the existence of the lexical pitch accent system. However, several questions remain for defining that categorical production, which leads to ideas for future research.

• The Journal of the Acoustical Society of Korea
• /
• v.20 no.3
• /
• pp.56-62
• /
• 2001

In this paper, we adopt a missing data theory to speech recognition. It can be used in order to maintain high performance of speech recognizer when the missing data occurs. In general, hidden Markov model (HMM) is used as a stochastic classifier for speech recognition task. Acoustic events are represented by continuous probability density function in continuous density HMM(CDHMM). The missing data theory has an advantage that can be easily applicable to this CDHMM. A marginalization method is used for processing missing data because it has small complexity and is easy to apply to automatic speech recognition (ASR). Also, a spectral subtraction is used for detecting missing data. If the difference between the energy of speech and that of background noise is below given threshold value, we determine that missing has occurred. We propose a new method that examines the reliability of detected missing data using voicing probability. The voicing probability is used to find voiced frames. It is used to process the missing data in voiced region that has more redundant information than consonants. The experimental results showed that our method improves performance than baseline system that uses spectral subtraction method only. In 452 words isolated word recognition experiment, the proposed method using the voicing probability reduced the average word error rate by 12％ in a typical noise situation.

• The Journal of the Acoustical Society of Korea
• /
• v.24 no.8
• /
• pp.498-504
• /
• 2005

We introduce a Korean speech recognition platform (ECHOS) developed for education and research Purposes. ECHOS lowers the entry barrier to speech recognition research and can be used as a reference engine by providing elementary speech recognition modules. It has an easy simple object-oriented architecture, implemented in the C++ language with the standard template library. The input of the ECHOS is digital speech data sampled at 8 or 16 kHz. Its output is the 1-best recognition result. N-best recognition results, and a word graph. The recognition engine is composed of MFCC/PLP feature extraction, HMM-based acoustic modeling, n-gram language modeling, finite state network (FSN)- and lexical tree-based search algorithms. It can handle various tasks from isolated word recognition to large vocabulary continuous speech recognition. We compare the performance of ECHOS and hidden Markov model toolkit (HTK) for validation. In an FSN-based task. ECHOS shows similar word accuracy while the recognition time is doubled because of object-oriented implementation. For a 8000-word continuous speech recognition task, using the lexical tree search algorithm different from the algorithm used in HTK, it increases the word error rate by $40\%$ relatively but reduces the recognition time to half.

• Phonetics and Speech Sciences
• /
• v.6 no.4
• /
• pp.53-65
• /
• 2014

The current research examines whether pitch contour shapes in North Kyungsang pitch accent contrasts provide a phonetic dimension for phonological discreteness in a mimicry task. Two pitch accent continua resynthesized were created for HH vs. HL and HH vs. LH. To confirm a phonetic dimension for accounting for pitch accent categories in North Kyungsang Korean, the mimicries of speakers of two dialects (i.e., North Kyungsang & South Cholla) were compared. One of the findings showed that, for North Kyungsang speakers, the range of mean f0 peak times was a phonetic dimension undergoing a continuous shift within a stimulus continuum for both HH vs. HL and HH vs. LH. On the other hand, for South Cholla speakers, there were no apparent shifts around categorical boundaries for either HH vs. HL or HH vs. LH. Regarding individual mimicries on f0 peak timing, there are many variations. For HH vs. LH, three North Kyungsang speakers showed a discrete pattern reflecting a shift in phonological categories, but for HH vs. HL, there was no such distinction showing a categorical shift, though there were statistically significant differences for two speakers. Interestingly, one of the North Kyungsang speakers showed a continuous phonetic dimension for both HH vs. HL and HH vs. LH. Lastly, the f0 valley timing did not exhibit a discrete or gradient phonetic dimension for speakers of either dialect. On the basis of these results, what is interesting is that the tonal target such as high tone in North Kyungsang pitch accent categories within the autosegmental-metrical (AM) theory may be realized within individual cognitive systems for representing the interaction of perception and production.

• Proceedings of the KSPS conference
• /
• 2005.11a
• /
• pp.215-218
• /
• 2005

This paper reports the current status of development of the Korean speech recognition platform (ECHOS). We implement new modules including ETSI feature extraction, backward search with trigram, and utterance verification. The ETSI feature extraction module is implemented by converting the public software to an object-oriented program. We show that trigram language modeling in the backward search pass reduces the word error rate from 23.5% to 22% on a large vocabulary continuous speech recognition task. We confirm the utterance verification module by examining word graphs with confidence score.

• The Journal of the Acoustical Society of Korea
• /
• v.19 no.6
• /
• pp.62-70
• /
• 2000

In usual language models(LMs), the probability has been estimated by selecting highly frequent words from a large text side database. However, in case of adopting LMs in a specific task, it is unnecessary to using the general method; constructing it from a large size tent, considering the various kinds of cost. In this paper, we propose a construction method of LMs using a small size text database in order to be used in specific tasks. The proposed method is efficient in increasing the low frequent words by applying same sentences iteratively, for it will robust the occurrence probability of words as well. We carried out continuous speech recognition(CSR) experiments on 200 sentences uttered by 3 speakers using LMs by iterative teaming(IL) in a air flight reservation task. The results indicated that the performance of CSR, using an IL applied LMs, shows an 20.4% increased recognition accuracy compared to those without it. This system, using the IL method, also shows an average of 13.4% higher recognition accuracy than the previous one, which uses context-free grammar(CFG), implying the effectiveness of it.

• MALSORI
• /
• no.50
• /
• pp.51-69
• /
• 2004

In this paper, we present a one-pass semi-dynamic network decoding framework that inherits both advantages of fast decoding speed from static network decoders and memory efficiency from dynamic network decoders. Our method is based on the novel language model network representation that is essentially of finite state machine (FSM). The static network derived from the language model network [1][2] is partitioned into smaller subnetworks which are static by nature or self-structured. The whole network is dynamically managed so that those subnetworks required for decoding are cached in memory. The network is near-minimized by applying the tail-sharing algorithm. Our decoder is evaluated on the 25k-word Korean broadcast news transcription task. In case of the search network itself, the network is reduced by 73.4% from the tail-sharing algorithm. Compared with the equivalent static network decoder, the semi-dynamic network decoder has increased at most 6% in decoding time while it can be flexibly adapted to the various memory configurations, giving the minimal usage of 37.6% of the complete network size.

• The Journal of the Acoustical Society of Korea
• /
• v.15 no.4
• /
• pp.24-31
• /
• 1996

In this paper, we present a speech translation system for hotel reservation, KT_STS(Korea Telecom Speech Translation System). KT-STS is a speech-to-speech translation system which translates a spoken utterance in Korean into one in Japanese. The system has been designed around the task of hotel reservation(dialogues between a Korean customer and a hotel reservation de나 in Japan). It consists of a Korean speech recognition system, a Korean-to-Japanese machine translation system and a korean speech synthesis system. The Korean speech recognition system is an HMM(Hidden Markov model)-based speaker-independent, continuous speech recognizer which can recognize about 300 word vocabularies. Bigram language model is used as a forward language model and dependency grammar is used for a backward language model. For machine translation, we use dependency grammar and direct transfer method. And Korean speech synthesizer uses the demiphones as a synthesis unit and the method of periodic waveform analysis and reallocation. KT-STS runs in nearly real time on the SPARC20 workstation with one TMS320C30 DSP board. We have achieved the word recognition rate of 94. 68% and the sentence recognition rate of 82.42% after the speech recognition tests. On Korean-to-Japanese translation tests, we achieved translation success rate of 100%. We had an international joint experiment in which our system was connected with another system developed by KDD in Japan using the leased line.

• Proceedings of the IEEK Conference
• /
• 2002.07a
• /
• pp.98-101
• /
• 2002

This paper reports a parallel architecture in a HMM based speech recognition system for a low-power LSI design. The proposed architecture calculates output probability of continuous HMM (CHMM) by using concurrent and pipeline processing. They enable to reduce memory access and have high computing efficiency. The novel point is the efficient use of register arrays that reduce memory access considerably compared with any conventional method. The implemented system can achieve a real time response with lower clock in a middle size vocabulary recognition task (100-1000 words) by using this technique.