• ETRI Journal
• /
• v.36 no.5
• /
• pp.714-720
• /
• 2014

In this paper, we propose the use of data-driven probabilistic utterance-level decision logic to improve Weighted Finite State Transducer (WFST)-based endpoint detection. In general, endpoint detection is dealt with using two cascaded decision processes. The first process is frame-level speech/non-speech classification based on statistical hypothesis testing, and the second process is a heuristic-knowledge-based utterance-level speech boundary decision. To handle these two processes within a unified framework, we propose a WFST-based approach. However, a WFST-based approach has the same limitations as conventional approaches in that the utterance-level decision is based on heuristic knowledge and the decision parameters are tuned sequentially. Therefore, to obtain decision knowledge from a speech corpus and optimize the parameters at the same time, we propose the use of data-driven probabilistic utterance-level decision logic. The proposed method reduces the average detection failure rate by about 14% for various noisy-speech corpora collected for an endpoint detection evaluation.

• Annual Conference on Human and Language Technology
• /
• 1995.10a
• /
• pp.181-187
• /
• 1995

한국어 형태소 해석기와 같은 한국어 정보 치리 시스템은 많은 전자 사진 검색 작업을 요구하기 때문에 전자 사전의 성능은 전체 시스템의 성능에 많은 영향을 미친다. 이에 본 논문은 적은 기억 장소를 차지하면서 탐색 속도가 빠른 Finite State Transducer(FST)를 이용한 전자 사전 구조를 제안한다. 제안된 전자 사진은 Deterministic Finite State Automata(DFA)로 표제어를 표현하고 DFA 상태수 최소화 알고리즘으로 모든 위치에 존재하는 중복된 상태를 제거하여 필요한 기억 장소가 적으며, FST를 일차원 배열에 매핑하고 탐색시 이 배열내에서의 상태 전이만으로 탐색을 하기 때문에 탐색 속도가 매우 빠르다. 또한 TRIE 구조에서와 같이 한번의 탐색으로 입력된 단어로 가능한 모든 표제어들을 찾아 줄 수 있다. 실험 결과 표제어 수가 증가하여도 FST를 이용한 전자 사전의 크기는 표제어 수에 비례하여 커지지 않고, 전자 사전 탐색 시간은 표제어 수에 영향을 받지 않으며, 약 237만 단어를 검색하는 실험에서 TRIE나 $B^+-Tree$구조를 사용한 전자 사전보다 빠름을 알 수 있었다.

• Journal of the Korean Society for Nondestructive Testing
• /
• v.32 no.3
• /
• pp.241-262
• /
• 2012

Some applications require structural health monitoring in inaccessible components. This paper presents a technique useful for Structural Health Monitoring of double wall structures, such as double wall steam pipes and double wall pressure vessels separated from an ultrasonic transducer by three layers. Detection has been demonstrated at distances in excess of one meter for a fixed transducer. The case presented here is for one of the layers, the middle layer, being a fluid. For certain transducer configurations the wave propagating in the fluid is a wave with low velocity and attenuation. The paper presents a model based on wave theory and finite element simulation; the experimental set-up and observations, and comparison between theory and experiment. The results provide a description of the technique, understanding of the phenomenon and its possible applications in Structural Health Monitoring.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.3
• /
• pp.335-341
• /
• 2022

In this paper, we adopt a transformer that shows remarkable performance in natural language processing as an acoustic model of hybrid speech recognition. The transformer acoustic model uses attention structures to process sequential data and shows high performance with low computational cost. This paper proposes a method to improve the performance of transformer AM by applying each of the four algorithms of sequence discriminative training, a weighted finite-state transducer (wFST)-based learning used in the existing DNN-HMM model. In addition, compared to the Cross Entropy (CE) learning method, sequence discriminative method shows 5 % of the relative Word Error Rate (WER).

• Smart Structures and Systems
• /
• v.17 no.6
• /
• pp.1055-1065
• /
• 2016

An anisotropic ultrasonic transducer is proposed for Lamb wave applications, such as passive damage or impact localization based on ultrasonic guided wave theory. This transducer is made from a PMNPT single crystal, and has different piezoelectric coefficients $d_{31}$ and $d_{32}$, which are the same for the conventional piezoelectric materials, such as Lead zirconate titanate (PZT). Different piezoelectric coefficients result in directionality of guided wave generated by this transducer, in other words, it is an anisotropic ultrasonic transducer. And thus, it has different sensitivity in comparison with conventional ultrasonic transducer. The anisotropic one can provide more information related to the direction when it is used as sensors. This paper first shows its detailed properties, including analytical formulae and finite elements simulations. Then, its application is described.

• Journal of KSNVE
• /
• v.6 no.1
• /
• pp.65-70
• /
• 1996

The sensor response of a piezoelectric transducer embedded in a fluid loaded structure is modeled using a hybrid numerical approach. The structure is excited by an obliquely incident acoustic wave. Finite element modeling in the structure and fluid surrounding the transducer region, is used and a plane wave representation is exploited to match the displacement field at the mathematical boundary. On this boundary, continuity of field derivatives is enforced by using a penalty factor and to further achieve transparency at the mathematical boundary, drilling degrees of freedom (d.o.f.) are introduced to ensure continuity of all derivatives. Numerical results are presented for the sensor response and it is found that the sensor at that location is not only non-intrusive but also sensitive to the characteristic of the structure.

• Journal of Advanced Marine Engineering and Technology
• /
• v.37 no.7
• /
• pp.760-770
• /
• 2013

Detecting grammatical errors from a text is a long-history application. In this paper, we compare the performance of two grammatical error detection techniques, which are implemented as a sub-module of an automated English scoring system. One is to use a full syntactic parser, which has not only grammatical rules but also extra-grammatical rules in order to detect syntactic errors while paring. The other one is to use a finite state machine which can identify an error covering a small range of an input. In order to compare the two approaches, grammatical errors are divided into three parts; the first one is grammatical error that can be handled by both approaches, and the second one is errors that can be handled by only a full parser, and the last one is errors that can be done only in a finite state machine. By doing this, we can figure out the strength and the weakness of each approach. The evaluation results show that a full parsing approach can detect more errors than a finite state machine can, while the accuracy of the former is lower than that of the latter. We can conclude that a full parser is suitable for detecting grammatical errors with a long distance dependency, whereas a finite state machine works well on sentences with multiple grammatical errors.

• Annual Conference on Human and Language Technology
• /
• 1999.10e
• /
• pp.128-134
• /
• 1999

기계번역이나 정보검색 등에 적용되는 자연언어처리기술에 있어서 구문분석은 매우 중요한 위치를 차지한다. 하지만, 문장의 길이가 증가함에 따라 구문분석의 복잡도는 크게 증가하게 된다. 이를 해결하기 위한 많은 노력 중에서 전처리기의 지원을 통해 구문분석기의 부담을 줄이려는 방법이 있다. 본 논문에서는 구문분석의 애매성과 복잡성을 감소시키기 위해 유한상태 변환기 (Finite-State-Transducer FSI)를 이용한 전처리기를 제안한다. 유한상태 변환기는 사전표현, 단어분할, 품사태깅 등에 널리 사용되어 왔는데, 본 논문에서는 유한상태 변환기를 이용하여 형태소 분석된 문장에서 시간표현 등의 제한된 표현들을 구문요소화하는 전처리기를 설계 및 구현하였다. 본 논문에서는 기계번역기에서의 구문분석기 뿐만 아니라 변환지식의 모듈화를 지원하기 위해 유한상태 변환기를 이용하여 시간표현 등의 부분적인 표현들을 번역하는 방법을 제안한다. 또한 유한상태 변환기의 편리한 작성을 위하여 유한상태 변환기 작성 지원도구를 구현하였다. 본 논문에서는 전처리기의 적용을 통해 구문분석기의 부담을 덜어 주며 기계번역기의 변환부분의 일부를 성공적으로 담당할 수 있음을 보여 준다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.6
• /
• pp.1692-1705
• /
• 2024

In this paper, we present a method that integrates a Grammar Transducer as an external language model to enhance the accuracy of the pre-trained Korean End-to-end (E2E) Automatic Speech Recognition (ASR) model. The E2E ASR model utilizes the Connectionist Temporal Classification (CTC) loss function to derive hypothesis sentences from input audio. However, this method reveals a limitation inherent in the CTC approach, as it fails to capture language information from transcript data directly. To overcome this limitation, we propose a fusion approach that combines a clause-level n-gram language model, transformed into a Weighted Finite-State Transducer (WFST), with the E2E ASR model. This approach enhances the model's accuracy and allows for domain adaptation using just additional text data, avoiding the need for further intensive training of the extensive pre-trained ASR model. This is particularly advantageous for Korean, characterized as a low-resource language, which confronts a significant challenge due to limited resources of speech data and available ASR models. Initially, we validate the efficacy of training the n-gram model at the clause-level by contrasting its inference accuracy with that of the E2E ASR model when merged with language models trained on smaller lexical units. We then demonstrate that our approach achieves enhanced domain adaptation accuracy compared to Shallow Fusion, a previously devised method for merging an external language model with an E2E ASR model without necessitating additional training.

• Transactions of the Korean Society for Noise and Vibration Engineering
• /
• v.19 no.12
• /
• pp.1347-1355
• /
• 2009

This paper demonstrates the validity of spectral element analysis for modeling the high-frequency dynamic behaviors of a beam with a surface-bonded piezoelectric wafer through a laboratory test. In the spectral element analysis, the high-frequency electro-mechanical interaction can be considered properly with relatively low computational cost compared to the finite element analysis. In the verification test, a cantilever beam with a surface-bonded piezoelectric wafer is forced to be in steady-state motion by exerting the harmonic driving voltage signal on the piezoelectric wafer. A laser scanning vibrometer is used to obtain the overall dynamic responses of the structure such as resonance frequencies, the associated mode shapes, and frequency response functions up to 20 kHz. Then, these dynamic responses from the test are compared to those computed by the spectral element analysis. A two-dimensional finite analysis is conducted to obtain the asymptotic solutions for the comparison purpose as well.