• Journal of KIISE
• /
• v.42 no.11
• /
• pp.1372-1379
• /
• 2015

This paper presents an analysis of feature extraction methods used for distinguishing the speech of patients with cleft palates and people with normal palates. This research is a basic study on the development of a software system for automatic recognition and restoration of speech disorders, in pursuit of improving the welfare of speech disabled persons. Monosyllable voice data for experiments were collected for three groups: normal speech, cleft palate speech, and simulated clef palate speech. The data consists of 14 basic Korean consonants, 5 complex consonants, and 7 vowels. Feature extractions are performed using three well-known methods: LPC, MFCC, and PLP. The pattern recognition process is executed using the acoustic model GMM. From our experiments, we concluded that the MFCC method is generally the most effective way to identify speech distortions. These results may contribute to the automatic detection and correction of the distorted speech of cleft palate patients, along with the development of an identification tool for levels of speech distortion.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.40 no.1
• /
• pp.74-81
• /
• 2003

In this study, we'd implemented a real time embedded speech recognition system that requires minimum memory size for speech recognition engine and DB. The word to be recognized consist of 40 commands used in a PCS phone and 10 digits. The speech data spoken by 15 male and 15 female speakers was recorded and analyzed by short time analysis method, which window size is 256. The LPC parameters of each frame were computed through Levinson-Burbin algorithm and they were transformed to Cepstrum parameters. Before the analysis, speech data should be processed by pre-emphasis that will remove the DC component in speech and emphasize high frequency band. Baum-Welch reestimation algorithm was used for the training of HMM. In test phone, we could get a recognition rate using likelihood method. We implemented an embedded system by porting the speech recognition engine on ARM core evaluation board. The overall recognition rate of this system was 95%, while the rate on 40 commands was 96% and that 10 digits was 94%.

• Proceedings of the KSPS conference
• /
• 1996.10a
• /
• pp.514-514
• /
• 1996

Igbo, a new Benue Congo language has a vowel harmony system which, like that of Akan, is based on the pharynx size or tongue root position. In this study we examine Igbo vowel harmony with particular reference to assimilatory patterns of vowels in different harmony sets. This is to gain some insight into the factors involved in Igbo vowel assimilation, and to establish to what extent reports on Akan vowel assimilation are validated in Igbo. Tokens of the eight phonemic vowels of Standard Igbo are recorded from three native speakers of Igbo. The vowels are acoustically investigated (using the LPC analysis of CSL) in individual lexical items and within carefully designed carrier phrases. The F1 and F2 values of the vowels are obtained as these formant values are generally useful in establishing the salient characteristics of vowels. Vowels from the harmony sets are juxtaposed in the carrier phrases to ascertain the extent of assimilation. Results of the investigation show that the F1 values, to a large extend, are enough to characterize these vowels. The (-Expanded) vowels have higher F1 values than their (+Expanded) counterpart. Where there is an overlap in F1 values for some vowels the F1 bandwidth values serve to distinguish between the vowels. The overlap often reported in Akan for /I/ and /e/ on the one hand and /${\mho}$/ and /o/ on the other is not validated in Igbo. While the F1 values for these pairs of vowels are quite similar for one of our speakers, there is an appreciable difference between the F1 values of these vowels for the other two speakers. There is however an overlap for /e/ and /o/ for one of the speakers. Assimilations are generally regressive across word boundaries. It is, however, necessary to point out that the general perceptual impression that one of the vowels completely assimilates to the other, is not borne out by our investigation. Most of our F1 and F2 values for the vowels in individual lexical items are altered in assimilations. This then suggests that assimilation involving these vowels is partial rather than complete. The emerging 'allophones' are acoustically similar to the (+Expanded) vowel involved in the assimilation, that is when vowels from different harmony sets are involved. We conclude that while assimilation of Igbo vowels involves some phonological considerations, phonetic factors appear to be permanent in deciding the final form of the vowels.

• Journal of the Korean Society of Laryngology, Phoniatrics and Logopedics
• /
• v.5 no.1
• /
• pp.44-58
• /
• 1994

The formants of the 9 Korean standard vowels(which used by the average people of Seoul, central-area of the Korean peninsula) were measured by analysis with the linear predictive coding(LPC) and fast Fourier transform(FFT). The author already had reported the constriction area for the Korean standard vowels, and with the existing data, the distance from glottis to the constriction area in the vocal tract of each vowel was newly measured with videovelopharyngograms and lateral Rontgenograms of the vocal tract. We correlated the formant frequencies with the distance from glottis to the constriction area of the vocal tract. Also we tried to correlate the formant frequencies with the position of tongue in the vocal tract which is divided into 2 categories : The position of tongue in oral cavity by the distance from imaginary palatal line to the highest point of tongue and the position in pharyngeal cavity by the distance from back of tongue to posterior pharyngeal wall. This study was performed with 10 adults(male : 5, female : 5) who spoke primary 9 Korean standard vowels. We had already reported that the Korean vowel [i], [e], $[{\varepsilon}]$ were articulated at hard palate level, [$\dot{+}$], [u] were at soft palate level, [$\wedge$] was at upper pharynx level and the [$\wedge$], [$\partial$], [a] in a previous article. Also we had noted that the significance of pharyngeal cavity in vowel articulation. From this study we have concluded that ; 1) The F$_1$ is related with the oral cavity articulated vowel [i, e, $\varepsilon$, $\dot{+}$, u]. 2) Within the oral cavity articulated vowel [i, e, $\varepsilon$, $\dot{+}$, u] and the upper pharynx articulated vowel [o], the F$_2$ is elevated when the diatance from glottis to the constriction area is longer. But within the lower pharynx articulated vowel [$\partial$, $\wedge$, a], the F$_2$ is elevated when the distance from glottis to the constriction area is shorter. 3) With the stronger tendency of back-vowel, the higher the elevation of the F$_1$ and F$_2$ frequencies. 4) The F$_3$ and F$_4$ showed no correaltion with the constriction area nor the position of tongue in the vocal tract 5) The parameter F$_2$- F$_1$, which is the difference between F$_2$ frequency and F$_1$ frequency showed an excellent indicator of differenciating the oral cavity articulated vowels from pharyngeal cavity articulated vowels. If the F$_2$-F$_1$ is less than about 600Hz which indicates the vowel is articulated in the pharyngeal cavity, and more than about 600Hz, which indicates that the vowel is articulated in the oral cavity.

• Journal of the Korea Society of Computer and Information
• /
• v.18 no.7
• /
• pp.23-35
• /
• 2013

This paper proposes an effective technique that is used to automatically extract feature vectors from vibration signals for fault classification systems. Conventional mel-frequency cepstral coefficients (MFCCs) are sensitive to noise of vibration signals, degrading classification accuracy. To solve this problem, this paper proposes spectral envelope cepstral coefficients (SECC) analysis, where a 4-step filter bank based on spectral envelopes of vibration signals is used: (1) a linear predictive coding (LPC) algorithm is used to specify spectral envelopes of all faulty vibration signals, (2) all envelopes are averaged to get general spectral shape, (3) a gradient descent method is used to find extremes of the average envelope and its frequencies, (4) a non-overlapped filter is used to have centers calculated from distances between valley frequencies of the envelope. This 4-step filter bank is then used in cepstral coefficients computation to extract feature vectors. Finally, a multi-layer support vector machine (MLSVM) with various sigma values uses these special parameters to identify faulty types of induction motors. Experimental results indicate that the proposed extraction method outperforms other feature extraction algorithms, yielding more than about 99.65% of classification accuracy.