• 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.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2006.05a
• /
• pp.37-40
• /
• 2006

계통분석을 하는 생물학자들은 관련된 분석대상에 대한 정보를 확보하여 비교분석하기 위해 NCBI 등으로부터 염기서열을 확보하여 아미노산 서열로 변환하는 작업을 수행하게 된다. 많은 서열 데이터에 대해서 데이터베이스로부터 데이터를 검색하고 이를 변환하는 작업을 순차적으로 분석자가 관여하여 작업하는 것이 현재 분석환경이다. 따라서 본 논문에서는 분석의 효율성을 향상시키기 위해, 관심서열의 등록번호(Accession Number) 리스트를 입력하면 해당 서열에 대항 정보를 NCBI로부터 웹로봇을 통해 자동으로 확보한 다음, 확보된 염기서열 전체를 아미노산 서열로 자동 변환하여 가장 긴 ORF(Open Reading Frame)을 추천해주기 위해 설계된 지능형 다중 염기서열 변환 도구에 대해서 소개한다.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1995.06a
• /
• pp.228-231
• /
• 1995

반연속 HMM 음성인식 시스템의 화자 적응 성능 향상을 위해 코드북 변환 알고리즘을 제안하였다. 기존의 화자 적응 알고리즘으로는 새로운 화자의 적응 데이터 특징의 분포와 HMM 모수의 사전밀도를 함께 고려하는 베이시안 화자적응 알고리즘이 있다. 그러나 새로운 화자의 특징분포와 코드북 사전 밀도의 차이가 큰 경우 적응 데이터와 코드북간의 잘못된 대응 관계를 얻을 수 있으며, 기준 코드북에 필요 이상으로 많은 코드워드가 존재하는 경우 적응된 코드북에도 불필요한 코드워드 들이 남아 인식 과정에 혼란을 줄 수 있다. 이 문제점을 해결하기 위하여 제안된 코드북 변환 알고리즘에서는 주파수 영역의 포만트 정보를 이용하였다. 화자 적응을 수행하기 앞서 코드북의 켑스트럼으로부터 포만트를 추출해 내고, 이들의 분포를 적응 화자의 포만트 분포와 일치되도록 변환시켜 주었다. 이 변환된 포만트들로부터 다시 켑스트럼을 구하여 변환된 코드북을 얻고 이를 화자 적응의 초기 코드북으로 사용하였다. 제안된 알고리즘을 이용하였을 경우 코드북과 적응 화자의 음성 간의 정확한 대응관계를 찾을 수 있었고, 불필요한 코드워드들이 인식 과정에서 사용되지 않도록 변환되어 인식률이 향상되는 것을 실험을 통해 확인하였다.

• Journal of Korea Multimedia Society
• /
• v.7 no.5
• /
• pp.658-667
• /
• 2004

For digital image compression, various fractal image coding schemes using the self-similarity of image have been studied extensively. This paper discusses the problem that occurs during the calculating process of adaptive orthogonal transformation and provides improvements of LIFS coding scheme using the transformation. This proposed scheme has a better performance than JPEG for a wide range of compression ratio. This research also proposes an image composition method consisting of all domains of the transformation. The results show that the arithmetic operation processes of the encoder and the decoder become much smaller even without the distortion of the coding performance.

• The KIPS Transactions:PartD
• /
• v.14D no.3 s.113
• /
• pp.273-284
• /
• 2007

RFID middleware systems collect and filter the RFID streaming data gathered continuously by numerous readers in order to process requests from applications. These requests are called continuous queries because they are kept on executing during certain periods. To enhance the performance of the middleware, it is required to build an index to process the continuous queries efficiently. Several approaches of building an index on not data records but queries, called Query Index, are proposed and widely used for evaluating continuous queries over streaming data. The EPCglobal proposed an Event Cycle Specification (ECSpec) model, which is a standard query interface for RFID applications. Continuous queries based on ECSpec consist of a large number of segments for representing the query conditions. The problem with using any of the existing query indexes on these continuous queries is that it takes a long time to build the index because it is necessary to insert a large number of segments into the index. To solve this problem, we propose an Aggregate Transformation that converts a group of segments into a compressed data which is representative of the segments. We compare the performance of a transformed index with the existing query indexes.

• Journal of Advanced Navigation Technology
• /
• v.24 no.1
• /
• pp.53-60
• /
• 2020

In this paper, We have studied the structure of a Telemetry Transmitter capable of transmitting variable data rates. This paper proposed a structure combining variable pre-modulation filter with cutoff characteristic with variable input sample rate converter. Variable pre-modulation filter has the same characteristics as pre-modulation filter and is converted to a constant sampling rate without structural changes according to the variable input data rate. We propose a software program that actively controls variable pre-modulation filter and variable input sample rate converter to respond to real-time changing data.

• Journal of the Korea Computer Industry Society
• /
• v.5 no.5
• /
• pp.737-746
• /
• 2004

Two-channel multiwavelet system is investigated for image compression application in this paper. Generally, multiwavelets are known for their superb capability of compressing non-stationary signals like voice. However, multivavelet system have a critical problem in processing and compressing image data due to mesh-grid visual artifacts. In our two-channel multiwavelet system we have investigated incorporation of pre and post filtering to the multiwavelet transform and compression system for alleviating those ingerent visual artifacts due to multiwavelet effect. In addition, to quantify the image data compression performance of proposed multiwavelet system, computer simulations have been performed using various image data. For bit allocation and quantization, the Lagrange multiplier technique considering data rate vs. distortion rate along with a nonlinear companding method are applied equallly to all systems considered, here. The simulation results have yielded 1 ~ 2 dB compression enhancement over the scalar savelet systems. If the more advanced compression methods like SPIHT and run-length channel coding were adopted for the proposed multiwavelet system, a much higher compression gain could be obtained.

• Journal of the Korea Society of Computer and Information
• /
• v.9 no.4 s.32
• /
• pp.5-12
• /
• 2004

The data transformation between XML and Relation Database is made through the Principle of mapping bewtween them. There are two ways to access SQL Server, one is to assign SQL query to URL and the other is to use template file. MS-SQL server takes advantage of OpenXML function to transform the results of executing SQL query into XML documents. That is, OpenXML first makes node tree and then transforms row set data of XML documents into XML data of relation type. In order to insert XML data into database data. data is extracted from parsing XML documents using sp_xml_preparedocument procedure, and then the document structure is mapped into tree structure and stored in a table of database. Consequently, Data transformation between XML and Relation Database is made through mapping bewtween them. This article proposes the principle of mapping between XML and Relation Database and then shows the implementation of transformation between them so that it introduces the possibilty of bringing the extension and efficiency of data and various effects.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.35 no.11
• /
• pp.1521-1528
• /
• 2011

We studied the material properties for reliability improvement in finite element analysis results for a nitrile butadiene rubber hub-bearing seal and for a carbon-filled rubber mount used in a vehicle. It was difficult to measure the material properties of hundreds of types of rubber for the mount design. Thus, we suggested that the engineering stressstrain relations from pure shear test data could be synthesized by using simple tension data and Poisson's ratio. We defined Poisson's ratio by using a function of principal stretches to synthesize the stress-strain relations for a pure shear test. A transformation of the pure shear data was applied to the experimental values to obtain the predicted results when the strain approaches 100%. In the finite element analysis for the contact force of a hub-bearing seal, the strain results that used the transformation of the pure shear data and simple tension data almost corresponded to the experimental values. Ogden constants were used to analyze.

• Journal of the Korean Geographical Society
• /
• v.41 no.2 s.113
• /
• pp.188-194
• /
• 2006

Three-dimensional topological data is essential for 3D modeling and application such as emergency management and 3D network analysis. This paper reviewed current 3D topological data model and developed a method to construct 3D topological node-relation data structure from 2D computer aided design (CAD) data. The method needed two steps with medial axis-transformation and topological node-relation algorithms. Using a medial-axis transformation algorithm, the first step is to extract skeleton from wall data that was drawn polygon or double line in a CAD data. The second step is to build a topological node-relation structure by converting rooms to nodes and the relations between rooms to links. So, links represent adjacency and connectivity between nodes (rooms). As a result, with the conversion method 3D topological data for micro-level sub-unit of each building can be easily constructed from CAD data that are commonly used to design a building as a blueprint.