• The Journal of the Acoustical Society of Korea
• /
• v.18 no.2
• /
• pp.3-10
• /
• 1999

This paper describes the development and evaluation of a Korean address input system employing automatic speech recognition technique as user interface for input Korean address. Address consists of cities, provinces and counties. The system works on a window 95 environment of personal computer with built-in soundcard. In the speech recognition part, the Continuous density Hidden Markov Model(CHMM) for making phoneme like units(PLUs) and One Pass Dynamic Programming(OPDP) algorithm is used for recognition. For address recognition, Finite State Automata(FSA) suitable for Korean address structure is constructed. To achieve an acceptable performance against the variation of speakers, microphones, and environmental noises, Maximum a posteriori(MAP) estimation is implemented in adaptation. And to improve the recognition speed, fast search method using variable pruning threshold is newly proposed. In the evaluation tests conducted for the 100 connected words uttered by 3 males the system showed above average 96.0% of recognition accuracy for connected words after adaption and recognition speed within 2 seconds, showing the effectiveness of the system.

• Phonetics and Speech Sciences
• /
• v.5 no.1
• /
• pp.47-54
• /
• 2013

The aim of this study was to analyze consonant articulation errors produced by 90 speakers with cerebral palsy (CP). Phonetic transcriptions were made for 37 single-word utterances containing 70 phonemes: 48 initial consonants and 22 final consonants. Errors of substitution, omission, and distortion were analyzed using a confusion matrix paradigm showing the visualization of error patterns. Results showed that substitution errors in initial and final consonants were most frequent, followed by omission and distortion. Consonant omission occurred more frequently on final consonants. In both initial and final consonants, the within-place errors were more prominent than the within-manner errors. The current results suggest that consonant confusion matrices for dysarthric speech may provide useful information for evaluating speech intelligibility and developing automatic speech recognition system of adults with CP associated dysarthria.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.5
• /
• pp.476-482
• /
• 2020

Sequence-to-sequence deep neural networks with attention mechanisms have shown superior performance across various domains, where the sizes of the input and the output sequences may differ. However, if the input sequences are much longer than the output sequences, and the characteristic of the input sequence changes within a single output token, the conventional attention mechanisms are inappropriate, because only a single context vector is used for each output token. In this paper, we propose a double-attention mechanism to handle this problem by using two context vectors that cover the left and the right parts of the input focus separately. The effectiveness of the proposed method is evaluated using speech recognition experiments on the TIMIT corpus.

• The Journal of the Acoustical Society of Korea
• /
• v.25 no.5
• /
• pp.213-221
• /
• 2006

Phone segmentation of speech waveform is especially important for concatenative text to speech synthesis which uses segmented corpora for the construction of synthetic units. because the quality of synthesized speech depends critically on the accuracy of the segmentation. In the beginning. the phone segmentation was manually performed. but it brings the huge effort and the large time delay. HMM-based approaches adopted from automatic speech recognition are most widely used for automatic segmentation in speech synthesis, providing a consistent and accurate phone labeling scheme. Even the HMM-based approach has been successful, it may locate a phone boundary at a different position than expected. In this paper. we categorized adjacent phoneme pairs and analyzed the mismatches between hand-labeled transcriptions and HMM-based labels. Then we described the dominant error patterns that must be improved for the speech synthesis. For the experiment. hand labeled standard Korean speech DB from ETRI was used as a reference DB. Time difference larger than 20ms between hand-labeled phoneme boundary and auto-aligned boundary is treated as an automatic segmentation error. Our experimental results from female speaker revealed that plosive-vowel, affricate-vowel and vowel-liquid pairs showed high accuracies, 99%, 99.5% and 99% respectively. But stop-nasal, stop-liquid and nasal-liquid pairs showed very low accuracies, 45%, 50% and 55%. And these from male speaker revealed similar tendency.

• Proceedings of the KSPS conference
• /
• 2003.05a
• /
• pp.168-171
• /
• 2003

This study is about modeling pronunciation dictionary necessary for PLU(phoneme like unit) level word recognition. The recognition of nonnative speakers' pronunciation enables an automatic diagnosis and an error detection which are the core of English pronunciation tutoring system. The above system needs two pronunciation dictionaries. One is for representing standard English pronunciation. The other is for representing Korean speakers' English Pronunciation. Both dictionaries are integrated to generate pronunciation networks for variants.

• The Journal of the Korea Contents Association
• /
• v.9 no.1
• /
• pp.139-146
• /
• 2009

This paper addresses a modified spectral subtraction schemes which is suitable to speech recognition under low signal-to-noise ratio (SNR) noisy environment such as the automatic speech recognition (ASR) system in car. The conventional spectral subtraction schemes rely on the SNR such that attenuation is imposed on that part of the spectrum that appears to have low SNR, and accentuation is made on that part of high SNR. However, such postulation is adequate for high SNR environment, it is grossly inadequate for low SNR scenarios such as that of car environment. Proposed methods focused specifically to low SNR noisy environment by using weighting function for enhancing speech dominant region in speech spectrum. Experimental results by using voice commands for car show the superior performance of the proposed method over conventional methods.

• The Korean Journal of Applied Statistics
• /
• v.36 no.1
• /
• pp.33-48
• /
• 2023

In this paper we consider automatic speech recognition (ASR) for Korean speech data in which elderly persons randomly speak a sequence of words such as animals and vegetables for one minute. Most of the speakers are over 60 years old and some of them are dementia patients. The goal is to compare deep-learning based ASR models for such data and to find models with good performance. ASR is a technology that can recognize spoken words and convert them into written text by computers. Recently, many deep-learning models with good performance have been developed for ASR. Training data for such models are mostly composed of the form of sentences. Furthermore, the speakers in the data should be able to pronounce accurately in most cases. However, in our data, most of the speakers are over the age of 60 and often have incorrect pronunciation. Also, it is Korean speech data in which speakers randomly say series of words, not sentences, for one minute. Therefore, pre-trained models based on typical training data may not be suitable for our data, and hence we train deep-learning based ASR models from scratch using our data. We also apply some data augmentation methods due to small data size.

• Journal of Information Processing Systems
• /
• v.11 no.1
• /
• pp.22-38
• /
• 2015

While Modern Standard Arabic is the formal spoken and written language of the Arab world; dialects are the major communication mode for everyday life. Therefore, identifying a speaker's dialect is critical in the Arabic-speaking world for speech processing tasks, such as automatic speech recognition or identification. In this paper, we examine two approaches that reduce the Universal Background Model (UBM) in the automatic dialect identification system across the five following Arabic Maghreb dialects: Moroccan, Tunisian, and 3 dialects of the western (Oranian), central (Algiersian), and eastern (Constantinian) regions of Algeria. We applied our approaches to the Maghreb dialect detection domain that contains a collection of 10-second utterances and we compared the performance precision gained against the dialect samples from a baseline GMM-UBM system and the ones from our own improved GMM-UBM system that uses a Reduced UBM algorithm. Our experiments show that our approaches significantly improve identification performance over purely acoustic features with an identification rate of 80.49%.

• Journal of the Korea Society of Computer and Information
• /
• v.17 no.10
• /
• pp.25-33
• /
• 2012

In this paper, we propose a rule-based model to correct errors in a speech recognition result in the mobile device environment. The proposed model considers the mobile device environment with limited resources such as processing time and memory, as follows. In order to minimize the error correction processing time, the proposed model removes some processing steps such as morphological analysis and the composition and decomposition of syllable. Also, the proposed model utilizes the longest match rule selection method to generate one error correction candidate per point, assumed that an error occurs. For the purpose of deploying memory resource, the proposed model uses neither the Eojeol dictionary nor the morphological analyzer, and stores a combined rule list without any classification. Considering the modification and maintenance of the proposed model, the error correction rules are automatically extracted from a training corpus. Experimental results show that the proposed model improves 5.27% on the precision and 5.60% on the recall based on Eojoel unit for the speech recognition result.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.4
• /
• pp.1538-1552
• /
• 2021

Machine learning models are vulnerable to adversarial examples generated by adding a deliberately designed perturbation to a benign sample. Particularly, for automatic speech recognition (ASR) system, a benign audio which sounds normal could be decoded as a harmful command due to potential adversarial attacks. In this paper, we focus on the countermeasures against audio adversarial examples. By analyzing the characteristics of ASR systems, we find that frame offsets with silence clip appended at the beginning of an audio can degenerate adversarial perturbations to normal noise. For various scenarios, we exploit frame offsets by different strategies such as defending, detecting and hybrid strategy. Compared with the previous methods, our proposed method can defense audio adversarial example in a simpler, more generic and efficient way. Evaluated on three state-of-the-arts adversarial attacks against different ASR systems respectively, the experimental results demonstrate that the proposed method can effectively improve the robustness of ASR systems.