Dalwon Jang

Implementation of a matching engine for a practical query-by-singing/humming system

This paper proposes a matching engine of a query-by-singing/humming (QbSH) system of which database is constructed from polyphonic recordings such as MP3 files. Use of the database makes the system more practical since it saves the trouble of gathering MIDI files. The pitch sequences transcribed from polyphonic recordings may have errors, and to reduce the influence of the errors, the matching engine uses chroma-scale representation, compensation, and asymmetric dynamic time warping. We propose the use of saturated distances, and it is verified that the distances perform better then generally-used absolute difference and squared difference. In our experiment, our QbSH system achieved mean reciprocal rank of 0.725 for 1000 singing/ humming queries when searching from a database of 28 hour audio data.

MyMusicShuffler: Mood-based music recommendation with the practical usage of brainwave signals

This paper proposes an automatic music service, the MyMusicShuffler, which recommends music based on received brain signals. This service is focused on eliminating the unnecessary hand interactions in multi-tasking environments. By analyzing brainwave signals, the application can select music which effectively reflects the emotional responses of the user in real time. This paper explains the implementation of the service, MyMusicShuffler, a mood-based music recommendation service which interacts with the mood status of user.

Music recommendation system based on usage history and automatic genre classification

The personalized music recommender supports the user-favorite songs stored in a huge music database. In order to predict only user-favorite songs, managing user preferences information and genre classification are necessary. In our study, a very short feature vector, obtained from low dimensional projection and already developed audio features, is used for music genre classification problem. We applied a distance metric learning algorithm in order to reduce the dimensionality of feature vector with a little performance degradation. We propose the system about the automatic management of the user preferences and genre classification in the personalized music system.

Distance combination for content identification system

This paper considers an algorithm which produces a distance function by combining distance functions for a finger-printing system, which identifies a query content by matching its fingerprint to the database (DB) fingerprint. To match finger-prints, recent audio and video fingerprinting systems commonly use a simple distance metric such as l2 distance, and output the information of DB fingerprint that measure the shortest distance to the fingerprint of the query. This paper considers the weighted sum among various combining methods, and the weights are determined by the learning process with a given set of training data, which consists of the fingerprint of the distorted and the original contents. By solving an optimization problem which reduces the fingerprint distance between similar contents and increases the fingerprint distance between dissimilar contents, weights are determined. In our experiments, the proposed algorithm is applied to a video fingerprinting system, and the experimental results shows that combined distance outperforms the conventional l2 distance and that out algorithm to determine weights is reasonable.

Research about relation of music preference and brain-wave

In our research, classification of music preference based on brain-wave is studied. We assume that there is a clear difference between brain-wave when hearing favorite music and it when hearing disgusting music, and we collect the brain-wave of human while hearing the music. And, there are two methods of collecting: one is separation of favorite/disgusting music clips, and the other is mixing of them. In this paper, the experimental results for two cases, which are obtained with very simple classification method, are presented.

Phoneme based realtime taboo words similarity browsing system of new words using multi-lingual taboo words databases in web environments

This paper proposes the taboo words similarity browsing system which is able to assist the naming process. The proposed system in this paper analyzes new texts - new words or sentences - into phonemes, compares their similarities of pronunciations between new words and taboo words from various language cultural areas. The system supports the effective visualization interfaces in order to visualize and analyze the comparison results of similarity with taboo words which can be a primary factor decreasing the image about the brands in the specific cultural area.

Comparison of Frequency-domain block FxLMS and FxNLMS

An active noise control (ANC) is commonly applied to filtered-x least mean square (FxLMS) algorithm. But control stability can be fallen by characteristics of primary sources. So, this paper is description of control stability research using FxLMS and filtered-x normalized least mean square (FxNLMS) algorithm.

Design and Implementation of Panoramic Vision System Based on MPEG-V

This paper proposes the panoramic vision system using MPEG-V specification. ISO (International Organization for Standardization) developed MPEG-V (Moving Picture Experts Group-Virtual world) for the standardization of image-making, which is defined many sensors and actuators. We extend MPEG-V metadata specification about RADAR detectors and multiple cameras to describe information around cars. In our implementation, the proposed usage scenarios and metadata schema are applied to panoramic vision system, which consist of 4 cameras and OpenCV. The designed panoramic vision system can be applied to IVI (In Vehicle Information) system or monitoring system, providing a situation environment information of the area behind the cars.

Very short feature vector for music genre classiciation based on distance metric lerning

In our study, a very short feature vector, obtained from low dimensional projection and already developed audio features, is used for music genre classification problem. A long feature vector based on the concatenation of various features is generally used in music genre classification system. Our objective is to find a short feature vector, and we applied a distance metric learning algorithm in order to reduce the dimensionality of feature vector with a little performance degradation. In our experiments based on two widely-used dataset, dimension reduction based on distance metric learning is very effective, and we can get over 80% of accuracy with only 10-dimensional feature vector.

Test of pitch extraction algorithms for query-by-singing/humming system

This paper presents the test results of the Query-by-singing/humming(QbSH) system, which are measured with varying the pitch extraction algorithm. The test is for verifying matching engine of our QbSH system of which database is constructed from polyphonic recordings such as MP3 files. For the test, we used 3 different pitch extraction algorithms, and the experimental results are obtained with our matching engine. From the results, we can conclude that our matching engine works well with any pitch extraction algorithm.