The 11th COE Postdoctoral and Doctoral Researchers
Technical Presentation

Program (20 mins each: 15 mins presentation and 5 mins discussion)

1. "Evaluation of Soft Silicone Type NAM Microphones"
   Yoshitaka Nakajima (Speech and Acoustics Processing Laboratory : D2)
   中島 淑貴 (音情報処理学講座 : D2)
   

   [Abstract]
   The bandwidth of the NAM microphone has improved by using soft silicon for the sound medium compared with stethoscopic type. With soft silicone type NAM microphones we can sample expanded target voice signal by 5-10dB, suppressing air conduction noise signal to low by the experiment of synchronous stereo sampling of air and flesh conducted voices at the same gain. The HMM recognition accuracy of NAM and BTOS has improved with soft silicon type NAM microphone compared with stethoscopic type. Aural comprehension test showed that accuracy of catching sentences of NAM and BTOS by 12 testee has improved with soft silicon type NAM microphone almost as high as that of air conducted voices. However, the catching rate of a meaningless word was extremely low.
   
   Top

2. "Evaluate user panels for pant operations from the perceptual viewpoint"
   Xiwei Liu (Systems and Control Laboratory : D1')
   劉 希未 (システム制御・管理講座 : D1')
   

   [Abstract]
   In a human supervisory control system established in an industrial plant, an operator interacts with many graphic panels in real time. The construction of effective human-computer interaction (HCI) is critical to the safety and reliability of plant operation. In this presentation, I'd like to introduce a model-based approach to graphic panel evaluation. The evaluation employs HCI models and a perceptual processor model in the human information processing. A HCI model contains all information of icons, process and configuration variables in a user panel. The perceptual processor model defines a visual field, eye movement, and perceptual performance. By scanning throughout a user panel with different model parameters, visual strengths of all graphic items are recorded. According to the results, we can improve the layout of the user panel from the viewpoint of perception.
   
   Top

3. "Blind Source Separation using SIMO-model-based ICA with Self-Generator for Initial Filter"
   Tomoya Takatani (Speech and Acoustics Processing Laboratory : D1')
   高谷 智哉 (音情報処理学講座 : D1')
   

   [Abstract]
   We address the blind decomposition problem of the binaural mixed audio signals to realize the audio augmented reality system. In order to solve the problem, we have proposed Single-Input Multiple-Output (SIMO)-model-based indepednet component analysis (ICA) (SIMO-ICA) algorithm which decompose the mixed observed signals into the SIMO-model-based signals. Here the term "SIMO" represents the specific transmission system in which the input is a single source signal and the outputs are its transmitted signals observed at multiple microphones.
   In this presentation, in order to improve the decomposition performance, we propose a novel blind separation method using SIMO-ICA with a self-generator (SG) for the initial filter. Although an attractive feature of SIMO-ICA is beneficial to the binaural sound separation, SIMO-ICA has a serious drawback in its high sensitivity to the initial settings of the separation filter. In the proposed method, the SG functions as thepreprocessor of SIMO-ICA, and it can provide a valid initial filter for SIMO-ICA. To evaluate its effectiveness, binaural sound separation experiments are carried out under a reverberant condition. The experimental results reveal that the separation performance of the proposed method is superior to those of conventional methods.
   
   
   Top

4. "USB/IP: A Transparent Device Sharing over IP Network"
   Takahiro Hirofuchi (Information Technology Center : D1')
   広渕 崇宏 (情報科学センター : D1')
   

   [Abstract]
   Personal computing with affordable computers and their peripheral devices becomes more popular. To use such devices more efficiently and improve their usability, people want to share surrounding peripheral devices between computers without any modification of their own computing environments. The recent device sharing technologies in the pervasive computing area are not sufficient for the peripheral devices of personal computers because these technologies do not provide the network-transparency for applications and device drivers. I propose USB/IP as a transparent device sharing over IP network. This novel device sharing is based on the modern sophisticated peripheral interfaces and their supports in operating systems. By the Virtual Host Controller Interface Driver we have implemented as a peripheral bus driver, users can share diverse devices over networks without any modification in existing operating systems and applications. The experiments show that USB/IP has fairly practical I/O performance for various USB devices, including isochronous ones. I also show the performance optimization criteria for the further improvements.
   
   
   Top

5. "A Proposal of information retrieval method based on TPO metadata"
   Ismail Arai (Information Technology Center : D1')
   新井 イスマイル (情報科学センター : D1')
   

   [Abstract]
   In order to search contents that reflect user's circumstantial information in the Internet, the search engines should make use of some information such as current time, user's location, user's schedule and so on. We propose an information retrieval method which is making the best use of user's and content's metadata based on TPO (Time, Position, Occasion). We implement a model of information retrieval system based on the proposed method, which consists of query generation part, matching contents part, and scoring matched results part. As a result of designing query generation part with discretion, we achieve to reduce user's overload of choosing metadata. And a user becomes possible to retry sorting the matching results.
   
   
   Top

6. "Learning An Anaphoricity Determination Model Combining Preceding and Local Contextual Information"
   Ryu Iida (Computational Linguistics Laboratory : D1')
   飯田 龍 (自然言語処理学講座 : D1')
   

   [Abstract]
   We propose a machine learning-based approach to determine the anaphoricity of noun phrases by combining two types of clues: (i) non-anaphoric training instances as local contextual information and (ii) the most likely antecedents for a given target noun phrase as larger contextaul information. We design a model to utilize these two types of information effectively. Our experiments on Japanese noun phrases and zero-pronouns show that with such improvements, our model outperforms earlier learning-based approaches.
   
   
   Top

NAIST Graduate School of Information Science