• Phonetics and Speech Sciences
• /
• v.12 no.3
• /
• pp.55-63
• /
• 2020

Voice conversion can be applied to various voice processing applications. It can also play an important role in data augmentation for speech recognition. The conventional method uses the architecture of voice conversion with speech synthesis, with Mel filter bank as the main parameter. Mel filter bank is well-suited for quick computation of neural networks but cannot be converted into a high-quality waveform without the aid of a vocoder. Further, it is not effective in terms of obtaining data for speech recognition. In this paper, we focus on performing voice-to-voice conversion using only the raw spectrum. We propose a deep learning model based on the transformer network, which quickly learns the voice conversion properties using an attention mechanism between source and target spectral components. The experiments were performed on TIDIGITS data, a series of numbers spoken by an English speaker. The conversion voices were evaluated for naturalness and similarity using mean opinion score (MOS) obtained from 30 participants. Our final results yielded 3.52±0.22 for naturalness and 3.89±0.19 for similarity.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.14 no.5
• /
• pp.811-816
• /
• 2019

This paper first proposes a technique to separate clean speech signals and mixed noise signals by using an independent vector analysis algorithm of frequency band for 4 channel speech source signals with a noise. An improved output speech signal from the proposed independent vector analysis algorithm is obtained by using the cross-correlation between the signal outputs from the frequency domain delay-sum beamforming and the output signals separated from the proposed independent vector analysis algorithm. In the experiments, the proposed algorithm improves the maximum SNRs of 10.90dB and the segmental SNRs of 10.02dB compared with the frequency domain delay-sum beamforming algorithm for the input mixed noise speeches with 0dB and -5dB SNRs including white noise, respectively. Therefore, it can be seen from this experiment and consideration that the speech quality of this proposed algorithm is improved compared to the frequency domain delay-sum beamforming algorithm.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1998.06c
• /
• pp.171-174
• /
• 1998

본 논문에서는 음성합성(speech synthesis) 및 부호화(coding) 시스템에 있어서 음원(voice source) 모델링에 관한 문제를 살펴보고자 한다. 기존의 음원 모델링 시스템이 가지고 있는 여러 문제들을 극복하고자 기저함수(basis function) 의 가중 합(weighted-sum)으로 음원을 모델링 하는 새로운 기법을 제안하고자 한다. 제안한 방법에서는 음원 파형(voice source waveform)을 적절히 표현하기 위해서 필터뱅크(filter bank)에 기초한 기저함수의 가중 합으로 나타낸다. 다양한 음원 특성을 효과적으로 나타내는 음원 파라미터를 구하기 위하여 EM(estimate maximize)에 기초한 구조에 관해 조사한다. 제안한 방법을 이용하여 다양한 유성음에 대해 실험을 수행하였다. 실험결과 제안한 추정(estimation) 방법 및 모델링 방법을 이용하면 기존의 방법에 비해 더 정확한 음원 파형을 추정할 수 있고, 다양한 음원 특성을 나타낼 수 있다. 또한 음성합성 및 부호화에서도 음성품질(voice quality)를 개선시킬 수 있으리라 기대된다.

• MALSORI
• /
• no.57
• /
• pp.85-97
• /
• 2006

This paper suggests an algorithm that can estimate the direction of the sound source in real time. The algorithm uses the time difference and sound intensity information among the recorded sound source by four microphones. Also, to deal with noise of robot itself, the Kalman filter is implemented. The proposed method can take shorter execution time than that of an existing algorithm to fit the real-time service robot. Also, using the Kalman filter, signal ratio relative to background noise, SNR, is approximately improved to 8 dB. And the estimation result of azimuth shows relatively small error within the range of ${\pm}7$ degree.

• MALSORI
• /
• no.48
• /
• pp.19-33
• /
• 2003

Sasang Constitutional Medicine, a branch of traditional Korean medicine, believes that the health of human beings can be promoted by taking advantage of the fact that people have different constitutions. It utilizes the characteristics in human voice to diagnose the constitution of the patients. This study aims at establishing the relationship between Sasang constitutions and their corresponding voice characteristics by investigating source-related variables. Voice recordings of 23 patients from three different constitutions were obtained whose constitutions had been already diagnosed by the experts in the fields. Fundamental frequency related variables (average pitch, maximum/minimum pitch, pitch range), phonation type, speaking tempo were measured and analyzed for each group. The phonation type seemed to be a possible candidate for a successful variable to determine constitution. No statistically significant relationship was manifested between other variables and constitutions. Despite its failure to firmly establish the relationship between voice and constitutions, the current study suggests that future research should include not only source-related variables

• MALSORI
• /
• no.48
• /
• pp.35-55
• /
• 2003

The aim of this study is to clarify the process of the sound changes of Japanese consonants (Dakuons) in the analysis of the transcriptions of Korean Source-Materials (i.e. Japanese textbooks for Korean and the records of travel in Japan) written in the 15-l8th centuries with the Korean writing system. Especially these records of travel in Japan are meaningful in that the process of change of Dakuon is shown in detail. The results are as follows. 1) In the 15th century, all Dakuons /g d z b/ had nasality. 2) The nasality of /z/ and /b/ disappeared in the 16th century. 3) The nasality of /d/ disappeared in the late 17th century.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.137.2-137
• /
• 2001

We have been developed an intelligent mobile robot with an artificial olfactory function to recognize odours and to track odour source location. This mobile robot also has been installed an engine for speech recognition and synthesis, and is controlled by wireless communication. An artificial olfactory system based on array of 7 gas sensors has been installed in the mobile robot for odour recognition, and 11 gas sensors also are located in the bottom of robot to track odour sources. 3 optical sensors are also included in the intelligent mobile robot, which is driven by 2 D.C. motors, for clash avoidance in a way of direction toward an odour source. Throughout the experimental trails, it is confirmed that the intelligent mobile robot is capable of not only the odour recognition using artificial neural network algorithm, but also the tracking odour source using the step-by-step approach method ...

• Proceedings of the KSPS conference
• /
• 2006.05a
• /
• pp.49-52
• /
• 2006

The performance of a sound source localization system degrades severely in reverberant and noisy environments. In addition, restriction on the distance between microphones, which is required by portable devices, also lower the system performance. This paper compares the sound source localization algorithms based on time delay of arrival, which are robust to reverberation and noises considering microphone sensor distance. As well, post filter which outputs maximum count time delay is adopted to increase the accuracy.

• Speech Sciences
• /
• v.5 no.1
• /
• pp.167-179
• /
• 1999

The hearing mechanism is a complicated system. Sound is generated by a source that sends out air pressure or power. The pressure or power makes the sound waves. These waves reach the eardrum, or tympanic membrane, which vibrates at a rate and magnitude proportional to the nature of the sound waves. The tympanic membrane transforms this vibration into the mechanical energy in the middle ear, which in turn converts it to the hydraulic energy in the fluid of the inner ear. The hydraulic energy stimulates the sensory cells of the inner ear which send neuroelectrical impulses to the central auditory nervous system. The passive perception of auditory information starts just here. The listener gives attention to the speech sound, differentiates the sound from background noise, and integrates his experience with similar sounds. The listener then puts all of these aspects of audition into the context of the moment to identify the nature of sound. This has a major role in human communication. This paper provides an overview of the nature and characteristics of sound, the structure and function of the auditory system, and the way in which sound is processed by the auditory system.

• Asia Pacific Journal of Corpus Research
• /
• v.4 no.2
• /
• pp.75-87
• /
• 2023

This paper presents a revised version of the vocabulary analyzer for self-reflection (VACSR), called VACSR v.2.0. The initial version of the VACSR automatically analyzes the occurrences and the level of vocabulary items in the transcribed texts, indicating the frequency, the unused vocabulary items, and those not belonging to either scale. However, it overlooked words with multiple parts of speech due to their identical headword representations. It also needed to provide more explanatory result tables from different corpora. VACSR v.2.0 overcomes the limitations of its predecessor. First, unlike VACSR v.1, VACSR v.2.0 distinguishes words that are different parts of speech by syntactic parsing using Stanza, an open-source Python library. It enables the categorization of the same lexical items with multiple parts of speech. Second, VACSR v.2.0 overcomes the limited clarity of VACSR v.1 by providing precise result output tables. The updated software compares the occurrence of vocabulary items included in classroom corpora for each level of the Common European Framework of Reference-Japan (CEFR-J) wordlist. A pilot study utilizing VACSR v.2.0 showed that, after converting two English classes taught by a preservice English teacher into corpora, the headwords used mostly corresponded to CEFR-J level A1. In practice, VACSR v.2.0 will promote users' reflection on their vocabulary usage and can be applied to teacher training.