Songrit Maneewongvatana

Encoding emotions in speech with the size code — A perceptual investigation

Our current understanding of how emotions are expressed in speech is still very limited. Part of the difficulty has been the lack of understanding of the underlying mechanisms. Here we report the findings of a somewhat unconventional investigation of emotional speech. Instead of looking for direct acoustic correlates of multiple emotions, we tested a specific theory, the size code hypothesis of emotional speech, about two emotions – anger and happiness. According to the hypothesis, anger and happiness are conveyed in speech by exaggerating or understating the body size of the speaker. In two studies consisting of six experiments, we synthesized vowels with a three-dimensional articulatory synthesizer with parameter manipulations derived from the size code hypothesis, and asked Thai listeners to judge the body size and emotion of the speaker. Vowels synthesized with a longer vocal tract and lower F0 were mostly heard as from a larger person if the length and F0 differences were stationary, but from an angry person if the vocal tract was dynamically lengthened and F0 was dynamically lowered. The opposite was true for the perception of small body size and happiness. These results provide preliminary support for the size code hypothesis. They also point to potential benefits of theory-driven investigations in emotion research.

A recommendation model for personalized book lists

In this study, we present a novel approach to recommend the personalized book lists for the university members. Our approach consists of clustering the university members into different clusters based on their recent circulation activities and discovering the interest patterns of members in the cluster. In the first step, we clustered members sharing the common interests to the same cluster by using K-means algorithm, after that we explored the possible interest patterns performed by members in each cluster by association rules. Finally, we provided the recommended book lists that satisfy their individual needs and interest patterns. A questionnaire survey was performed to evaluate the accuracy satisfaction of predicting the satisfy book list to an individual. The evaluation results reveal the possibility of using circulation activity history to predict the current interest of an individual member and construct the personalized book lists that satisfy their interests.

Privacy preserving decision tree in multi party environment

Recently, there have been increasing interests on how to preserve the privacy in data mining when source of data are distributed across multi parties. In this paper, we focus on the privacy preserving on decision tree in multi party environment when data are vertically partitioned. We propose novel private decision tree algorithms applied to building and classification stages. The main advantage of our work over the existing ones is that each party cannot use the public decision tree to infer the other’s private data. With our algorithms, the communication cost during tree building stage is reduced compared to existing methods and the number of involving parties could be extended to be more than two parties.

Shape Control of a Hyper-Redundant Arm for Planar Object Manipulation

This paper discusses a method for controlling a hyper-redundant arm to manipulate an object on a plane. The hyper-redundant arm can perform simple whole-arm manipulation by coiling or wrapping around the object and then pulling the object toward the goal position. The process of object manipulation can be separated into two steps: encircling the object and transporting the object. In the process of encircling the object, the arm is controlled by a set of virtual constraints that guide the arm to reach around the object and encircle it, keeping the arm within a specified bound to ensure the circular shape around the object. In the process of transporting the object, a simplified desired shape is generated from a Bézier curve according to a given goal position and the arm geometry. Then, the gradient descent method is used to update the joint angles of the arm at each step to move the arm toward the desired shape until the object reaches its target position. The proposed method has been tested in both simulation and real experiments.

A similarity model for bibliographic records using subject headings

We propose a model for measuring similarity between bibliographic records based on their keywords or subject headings whose relationship can be represented in form of the semantic hierarchical network. In our work, the level of each subject heading in the semantic hierarchical network plays a major role to determine keywords infuence. Highly specifc subject headings represented as leaf nodes of the model are better indicators to explain the content of a bibliographic record than the more general ones. The proposed algorithm calculates the level of similarity between records by considering path from one subject heading to another contained in the records. With this algorithm, related bibliographic records do not necessary to have exactly the same subject heading. The implication of this work can be applied to various types of document classifcation whose set of tags or keywords are organized in a semantic hierarchical network.

Semantic personal image classification by energy expenditure

Semantic personal image classification is an attention problem in multimedia image retrieval. In our previous work [Chinpanchana, S et al., 2004], we classified semantic images into business, leisure, and sport categories by integrating the frequency pattern relationships between body parts and objects. However, the accuracy mainly depends on their objects. In the images that have high semantic complexities, the body movement play important solve on the meaning of image. In this paper, we present a new model to achieve more effective classifier called an energy expenditure model (EE). The EE model is based on the concept that human subjects in different classes of images are likely to spend different amounts of energy. The angular position and flexion forces are related into each body part. Experimental results show that the EE a can achieve an improvement of semantic images.

Using association rules to identify root causes of CRD in broilers

Chronic Respiratory Disease (CRD) in broilers poses some management challenges for the poultry meat industry since the disease discovery often comes too late. Raw meat from infected chickens must be excluded from being processed into cooked poultry product. Many factors from raising process are considered to contribute to such disease. Thus, finding root causes of CRD would help the farmers to have better control and prevent disease and also help the processing facilities to have a better estimation of raw meat amount. In this paper, we apply associate rules technique to identify the root causes of CRD. Possible factors were first identified by skilled veterinarians. Then collected data from several farms were used to find sets of factors that could potentially be the root causes. From the results we can identify set of general factors that are common root causes on all farms as well as sets of farm-dependent factors.

Improving performance of asynchronous BCI by using a collection of overlapping sub window models

Asynchronous Brain Computer Interfaces (BCI) have become an interesting topic in the present days because they provide simulation of realistic usage of BCI. For asynchronous BCI, the computer has to discriminate not only differences among various imaginary tasks but also detect relax periods. Since the training phase for building a classification model is still synchronous (cue-based), the main challenge is to classify the EEG signal continuously with good accuracy on asynchronous (uncue-based). This paper addresses achieving better performance by using a collection of overlapping sub windows models. A model is referred to a primitive classification model which consists of common spatial patterns (CSP) with linear discriminant analysis (LDA). Each primitive model was trained with the corresponding sub window indexes. We had 3 collections of models: task1 vs. task2, task1 vs. relax, and task2 vs. relax. These binary classification results were then fused together with Mahalanobis distance to gain better performance. The results were measured by mean square error (MSE), and their performance is better compared to the primitive model. Furthermore, the results on the test set were comparable to the 3 leading scores of BCI Competition IV dataset 1.

Identification of K-Tolerance Regulatory Modules in Time Series Gene Expression Data Using a Biclustering Algorithm

Nowadays, biclustering problem is still an intractable problem. But in time series expression data, the clusters can be limited those with contiguous columns. This restriction makes biclustering problem to be tractable problem. However existing contiguous column biclustering algorithm can only find the biclusters which have the same value for each column in biclusters without error tolerance. This characteristic leads the algorithm to overlook some patterns in its clustering process. We propose a suffix tree based algorithm that allows biclusters to have inconsistencies in at most k contiguous column. This can reveals previously undiscoverable biclusters. Our algorithm still has tractable run time with this additional feature.

Identification of Shifting Regulatory Modules in Time Series Gene Expression Data Using a Linear Time Biclustering Algorithm

— Since standard biclustering problem was defined, the problem is known to be NP-hard. However in analyzing time series expression data, we can restrict the problem with the trait of data that represented the contiguous columns, which corresponded to coherent expression patterns. With this restriction included, the problem considered to be tractable problem. We propose an algorithm to find and report all maximal contiguous column coherent biclusters with the shifting input included in time linear within the size of expression matrix multiplied by the size of the shifting window.