Sukanya Phongsuphap

Gender Identification from Thai Speech Signal Using a Neural Network

This paper proposes a method for identifying a gender by using a Thai spoken syllable with the Average Magnitude Difference Function (AMDF) and a neural network (NN). The AMDF is applied to extracting pitch contour from a syllable. Then the NN uses the pitch contour to identify a gender. Experiments are carried out to evaluate the effects of Thai tones and syllable parts on the gender classification performance. By using a whole syllable, the average correct classification rate of 98.5% is achieved. While using parts of a syllable, the first half part gives the highest accuracy of 99.5%, followed by the middle and the last parts with the accuracies of 96.5% and 95.5%, respectively. The results indicate that the proposed method using pitch contour from any tones of the first half of a Thai spoken syllable or a whole Thai spoken syllable with the NN is efficient for identifying a gender.

Marginal noise removal for scanned document images by projection profile based method

Significant documents or old scarce books have been saved in digital forms by scanning. However, some noises may occur in a scanned image such as marginal noises. The marginal noises usually appear as dark regions around the margin of the scanned documents. In this paper, we propose a method for removing marginal noise by analyzing image histogram or projection profile of the document image. The method consists of the following main steps: Creating image profile; Smoothing the profile by using moving average technique; Dividing the projection profile of the document image into three regions: left, middle, and right; Detecting marginal noises by finding the highest peaks of the projection profile in the corresponding regions; Removing the detected marginal noises by replacing the regions of marginal noises with white pixels. We had performed experiments on 50 input digital document images. The proposed method can detect and remove the marginal noises with a high average accuracy rate of 93.8% which is significantly higher than that of the tradition method, 77.5%, using the original projection profile.

Approximation method for interpolating affine parameters in doubling frame rate

The quality of the frame rate up-conversion depends on the accuracy of the motion parameters. 2D affine model describes the motion better than the popular 2D translation model; however, its parameters are difficult to interpolate. In this paper, the approximation method for interpolating the mid-point of 2D affine parameters is proposed. The affine transform is approximated as the 2D rigid transformation where the target x and y axes are not orthogonal. Trigonometry relationship is applied to interpolate rotational parameters. Conventional linear interpolation is applied to interpolate translational parameters. For video sequences containing complex motion, the reconstruction from the approximated affine parameters gave higher PSNR and visually better result than the one from the interpolated 2D translation parameters.

Two-Stage Gender Identification Using Pitch Frequencies, MFCCs and HMMs

This paper proposes a two-stage method, which can identify the gender of a speaker from spoken syllables that have different tones, using: an average pitch frequency, MFCC-based features (Mel-Frequency Cepstral Coefficients), and Hidden Markov Models (HMM). The method can be divided into 2 stages. At the first stage, an average pitch frequency of each speaker was used to classify the gender. Still, a number of ambiguous speakers who were not clearly classified at the first stage were then forwarded to the second stage. At the second stage, gender identification using: MFCC features, phoneme acoustic models for females and males, and grammar for gender recognition was applied. The experimental results show that the proposed method achieved a high correct gender identification rate of 98.92%, which was higher than the conventional method using an average pitch frequency and a threshold. The proposed method is also more accurate than the Artificial Neural Network (ANN) with pitch frequencies. The results indicate that the proposed method is a practical and efficient way to identify gender.

Evaluation of responsiveness of health systems using fuzzy-based technique

This paper proposes a method for evaluating responsiveness of health systems. The method is based on a fuzzy model, which can tackle uncertainty of survey data, and perform corresponding to the way that human being makes decisions and adjustments. To measure responsiveness of health systems, we have defined five fuzzy sets for two input variables: score of direct experience of using health service and score of anchoring vignette, and five fuzzy sets for one output variable: responsiveness score which is defined as the difference between score of direct experience of using health service and score of vignette. The twenty-five fuzzy rules are derived from the analysis of input and output variables association. Mamdani style inference technique is used to compute a crisp value of average responsiveness score for each component of health systems, and the overall average responsiveness score is computed by using the weight average method. The data of seven components based on WHO framework were collected from 4,446 outpatients of three schemes of health care systems in Thailand consisting of Civil Servant Medical Benefit Scheme (CSMBS), Social Security Scheme (SSS), and Universal Coverage Scheme (UCS). Results showed that CSMBS got the highest average responsiveness score followed by SSS which got a slightly higher average responsiveness score than UCS, but there are some variations in each of seven components. The proposed method of responsiveness evaluation can provide concise information both in terms of quantitative and qualitative measures, which can be used as a policy implication to assist government and health system policy makers in improving and providing the more suitable heath care services.

Medical document reader on Android smartphone

This paper presents a method for reading medical documents by using an Android smartphone. We have used techniques based on the Tesseract OCR Engine to extract the text content from medical document images such as a physical examination report. The following factors related to the document are considered: character font, text block size, and distance between the document and the camera on the phone. Based on experimental results, we found that among three character fonts (Angsana New, Calibri, and Tahoma), Calibri and Tahoma gave very high average accuracies (greater than 90%) for both character recognition and word recognition, but Angsana New gave quite a lower accuracy, about 75%. For the optimal distance between the document and the smartphone, the recommended distance is from 12 cm. to 15 cm. for a document block size of 21 × 3, 13 × 10, 12 × 8, or 10 × 13 cm2.

Frame reconstruction by double motion estimation of affine parameters

In video frame rate conversion, unavailable frames are reconstructed from existing frames based on interpolated motion vectors. Due to an error in estimating motion vectors and an occlusion effect, it is possible that intensity information is not provided to every pixel in the reconstructed frame. Pixels whose intensity is missing become holes in the reconstructed frame. This paper proposes the reconstruction without distinct holes by double motion vector estimation. Motion is described by affine model instead of the conventional 2D translation model. The proposed method applies block based motion estimation. It is divided into 2 stages: frame reconstruction and hole filling. In the reconstruction stage, a new frame is reconstructed by moving blocks in previous frames according to their interpolated motion vectors. Holes in the newly reconstructed frame are then filled with the corresponding pixels in the subsequent frame. The proposed method searches for the corresponding pixels by looking for the region in the subsequent frames that are similar to the surrounding region of each hole. The computation cost in the proposed method is low because the motion estimation is simple and applied only twice. It does not use any complex processes to fill holes. Experiments on three standard video sequences showed that our proposed method reconstructed frames in complex video sequence without noticeable holes.