• Proceedings of the Acoustical Society of Korea Conference
• /
• 1995.06a
• /
• pp.184-187
• /
• 1995

This paper proposes an algorithm for machine recognition of phonemes in continuous speech. The proposed algorithm is static strategy neural network. The algorithm uses, at the stage of training neuron, features such as PARCOR coefficient and auditory-like perceptual liner prediction . These features are extracted from speech samples selected by a sliding 25.6msec windows with s sliding gap being 3 msec long, then interleaved and summed up to 7 sets of parmeters covering 171 msec worth of speech for use of neural inputs. Perfomances are compared when either PARCOR or auditory-like PLP is included in the feture set.

• Journal of KIISE:Software and Applications
• /
• v.35 no.6
• /
• pp.374-385
• /
• 2008

Speech act tagging is an important step in various dialogue applications, which recognizes speaker's intentions expressed in natural language utterances. Previous approaches such as rule-based and statistics-based methods utilize the speech acts of previous utterances and sentence features of the current utterance. This paper proposes a method that determines speech acts of the current utterance using the speech acts of the following utterances as well as previous ones. Using the features of following utterances yields the accuracy 95.27%, improving previous methods by 3.65%. Moreover, sentence features of the previous utterances are employed to maximally utilize the information available to the current utterance. By applying the proper probability model for each speech act, final accuracy of 97.97% is achieved.

• Journal of KIISE:Software and Applications
• /
• v.30 no.3_4
• /
• pp.278-284
• /
• 2003

Speech act is the speaker's intentions indicated through utterances. It is important for understanding natural language dialogues and generating responses. This paper proposes the method of two stage that increases the performance of the korean speech act decision. The first stage is to select features from the part of speech results in sentence and from the context that uses previous speech acts. We use x$^2$ statistics(CHI) for selecting features that have showed high performance in text categorization. The second stage is to determine speech act with selected features and Neural Network. The proposed method shows the possibility of automatic speech act decision using only POS results, makes good performance by using the higher informative features and speed up by decreasing the number of features. We tested the system using our proposed method in Korean dialogue corpus transcribed from recording in real fields, and this corpus consists of 10,285 utterances and 17 speech acts. We trained it with 8,349 utterances and have test it with 1,936 utterances, obtained the correct speech act for 1,709 utterances(88.3%). This result is about 8% higher accuracy than without selecting features.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.8
• /
• pp.365-370
• /
• 2023

This paper aims to compare the performance of speech data classification into two groups, adult and elderly, based on the acoustic and linguistic characteristics that change due to aging, such as changes in respiratory patterns, phonation, pitch, frequency, and language expression ability. For acoustic features we used attributes related to the frequency, amplitude, and spectrum of speech voices. As for linguistic features, we extracted hidden state vector representations containing contextual information from the transcription of speech utterances using KoBERT, a Korean pre-trained language model that has shown excellent performance in natural language processing tasks. The classification performance of each model trained based on acoustic and linguistic features was evaluated, and the F1 scores of each model for the two classes, adult and elderly, were examined after address the class imbalance problem by down-sampling. The experimental results showed that using linguistic features provided better performance for classifying adult and elderly than using acoustic features, and even when the class proportions were equal, the classification performance for adult was higher than that for elderly.

• The KIPS Transactions:PartB
• /
• v.12B no.3 s.99
• /
• pp.365-368
• /
• 2005

We propose a speech act analysis method for Korean dialogue using Support Vector Machines (SVM). We use a lexical form of a word, its part of speech (POS) tags, and bigrams of POS tags as sentence features and the contexts of the previous utterance as context features. We select informative features by Chi square statistics. After training SVM with the selected features, SVM classifiers determine the speech act of each utterance. In experiment, we acquired overall $90.54\%$ of accuracy with dialogue corpus for hotel reservation domain.

• Proceedings of the KSPS conference
• /
• 2006.11a
• /
• pp.127-130
• /
• 2006

The purpose of this study is to examine the difference in acoustic features between Young Voices and Aged Voices, which are actually come from the same age group. The 12 female subjects in their thirties were participated and recorded their sustained vowel /a/, connected speech, and reading. Their voices were divided into Younger Voices and Aged Voices, which means voices sound like younger person and sound like in their age or more aged ones. Praat 4.4.22 was used to record and analyze their acoustic features like Fo, SFF, Jitter, Shimmer, HNR, Pitch-range. And the six female listeners guessed the subjects' age and judged whether they sound younger or as like their actual age. We used the Independent t-Test to find the significant difference between those two groups' acoustic features. The result shows a significant difference in Fo, SFF. The above and the previous studies tell us the group who sounds like younger or baby like voice has the similar acoustic features of actually young people.

• Journal of Internet Computing and Services
• /
• v.2 no.1
• /
• pp.97-106
• /
• 2001

In this paper, we design and implement complement based speech training system for voice disorder. The system consists of three level of training: precedent training, training for speech apprehension and training for speech enhancement. To analyze speech of voice disorder, we extracted speech features as loudness, amplitude, pitch using digital signal processing technique. Extracted features are converted to graphic interface for visual feedback of speech by the system.

• Phonetics and Speech Sciences
• /
• v.6 no.4
• /
• pp.79-90
• /
• 2014

An intonation curve stylization was used for an acoustical analysis of English speech. For the analysis, acoustical feature values were extracted from 1,848 utterances produced with normal and fast speech rate by 28 (12 women and 16 men) native speakers of English. Men are found to speak faster than women at normal speech rate but no difference is found between genders at fast speech rate. Analysis of pitch point features has it that fast speech has greater Pt (pitch point movement time), Pr (pitch point pitch range), and Pd (pitch point distance) but smaller Ps (pitch point slope) than normal speech. Men show greater Pt, Pr, and Pd than women. Analysis of sentence level features reveals that fast speech has smaller Sr (sentence level pitch range), Sd (sentence duration), and Max (maximum pitch) but greater Ss (sentence slope) than normal speech. Women show greater Sr, Ss, Sp (pitch difference between the first pitch point and the last), Sd, MaxNr (normalized Max), and MinNr (normalized Min) than men. As speech rate increases, women speak with greater Ss and Sr than men.

• The Journal of the Acoustical Society of Korea
• /
• v.24 no.7
• /
• pp.416-422
• /
• 2005

Speech recognition in digital communication systems has very low performance due to the spectral distortion caused by speech codecs. In this paper, the spectral distortion by speech codecs is analyzed and a pre-processing method which compensates for the spectral distortion is proposed for performance enhancement of speech recognition. Three standard speech codecs. IS-127 EVRC. ITU G.729 CS-ACELP and IS-96 QCELP. are considered for algorithm development and evaluation, and a single method which can be applied commonly to all codecs is developed. The performance of the proposed method is evaluated for three codecs, and by using the speech features extracted from the compensated spectrum. the recognition rate is improved by the maximum of $15.6\%$ compared with that using the degraded speech features.

• ETRI Journal
• /
• v.28 no.4
• /
• pp.502-505
• /
• 2006

A new class-based histogram equalization method is proposed for robust speech recognition. The proposed method aims at not only compensating the acoustic mismatch between training and test environments, but also at reducing the discrepancy between the phonetic distributions of training and test speech data. The algorithm utilizes multiple class-specific reference and test cumulative distribution functions, classifies the noisy test features into their corresponding classes, and equalizes the features by using their corresponding class-specific reference and test distributions. Experiments on the Aurora 2 database proved the effectiveness of the proposed method by reducing relative errors by 18.74%, 17.52%, and 23.45% over the conventional histogram equalization method and by 59.43%, 66.00%, and 50.50% over mel-cepstral-based features for test sets A, B, and C, respectively.