Speech audiometry assesses hearing disorders, typically relies on audiologists, making the process subjective and requiring in-person evaluation.In this paper, we introduce SylPh, a novel automatic syllable-level mispronunciation detection and diagnosis (MDD) model that generalizes across open-set syllables while also offering phonemic analysis.To capture a wide range of mispronunciation patterns, we construct positive and pseudo-negative bags to extract in-distribution and out-ofdistribution features from input audio.Our model aligns audio features with adaptive text embeddings using a contrastive objective, dynamically adjusting decision boundaries for each syllable within a single model.Extensive experiments on a largescale dataset demonstrate its effectiveness in both closed-set and open-set syllables.Notably, despite training only on syllablelevel labels, the Sylph has the capability to localize phonemelevel abnormalities, providing detailed diagnostic insights.

This paper presents our contributions to the Speech Emotion Recognition in Naturalistic Conditions (SERNC) Challenge, where we address categorical emotion recognition and emotional attribute prediction. To handle the complexities of natural speech, including intra- and inter-subject variability, we propose Multi-level Acoustic-Textual Emotion Representation (MATER), a novel hierarchical framework that integrates acoustic and textual features at the word, utterance, and embedding levels. By fusing low-level lexical and acoustic cues with high-level contextualized representations, MATER effectively captures both fine-grained prosodic variations and semantic nuances. Additionally, we introduce an uncertainty-aware ensemble strategy to mitigate annotator inconsistencies, improving robustness in ambiguous emotional expressions. MATER ranks fourth in both tasks with a Macro-F1 of 41.01% and an average CCC of 0.5928, securing second place in valence prediction with an impressive CCC of 0.6941.

Detecting early signs of hearing impairment may help prevent age-related cognitive decline and associated disorders. However, conventional methods like pure tone audiometry are often subjective and insufficient for early detection. In this paper, we propose a novel, subject-agnostic approach for automatically detecting hearing impairment by localizing abnormal speech patterns during word recognition tests. To achieve this, we collected an audio dataset stimulated by the standard word list that is commonly employed in clinical assessments of hearing loss. This dataset is segmented into phoneme-level audio, where we then extract and model the acoustic features exclusively from normal speech patterns in a semi-supervised setting. For inference, we compute anomaly scores at the phoneme level by comparing them with these learned normal patterns, enabling precise localization of abnormalities within speech. Our method demonstrates a high accuracy rate of 70.73% in detecting six levels of hearing impairment severity, showcasing its potential for assisting in early assessment.

This paper proposes a novel method for Alzheimer’s disease detection and MMSE prediction using a complementary and simultaneous ensemble (CONSEN) algorithm based on multilingual spontaneous speech. We define pause and intervention of speech to form disfluency features, as well as several acoustic features to train generalized models. With the help of the proposed CONSEN algorithm, our model achieves the best performance of 86.69% for AD detection and 3.727 RMSE for MMSE prediction, which is placed first rank in both tasks in ICASSP Signal Processing Grand Challenge: ADReSS-M Challenge 2023.

Research in social science has shown that the mobile phone users pay less attention to their surroundings, which exposes them to various hazards such as collisions with vehicles than other pedestrians. In this paper, we propose a novel handheld device that assists mobile phone users to walk more safely outdoors. The proposed system is implemented on a smart phone and uses its back camera to detect the current outdoor context, e.g. traffic intersections, roadways, and sidewalks, finally alerts the user of unsafe situations using sound and vibration from the phone. The outdoor context awareness is performed by three steps: preprocessing, feature extraction, and context recognition. First, it improves the image contrast while removing image noise, and then it extracts the color and texture descriptors from each pixel. Next, each pixel is classified as an intersection, sidewalk, or roadway using a support vector machine-based classifier. Then, to support the real-time performance on the smart phone, a multi-scale classification is applied to input image, where the coarse layer first discriminates the boundary pixels from the background and the fine layer categorizes the boundary pixels as sidewalk, roadway, or intersection. In order to demonstrate the effectiveness of the proposed method, some real-world experiments were performed, then the results showed that the proposed system has the accuracy of above 98% at the various environments. © Copyright 2016 by SCITEPRESS - Science and Technology Publications, Lda. All rights reserved.

In this paper, an online rating game called Image-Battle is developed to build the ground truth dataset for image sentiment analysis. Our goal is to provide more interesting and intuitive interface to users and to collect the images with more correct sentiment scores despite of less human intervention. For this, two schemes are designed: 1) a pair-wise competition and 2) a ranking algorithm based on visual link analysis. First, the system shows two images and asks the user which image is closer to a given sentiment. Thereafter, the ranking algorithm assigns the sentiment scores to the images based on all the competition results: the main idea is to give higher score to images that win more or to the images that beat images with high scores. To evaluate the proposed system, it was used to collect ground truth for 30,000 Photo.net images, each of which was labeled by six emotions. The ground truth was used to develop the sentiment recognition system, and its result was compared with that of other rating system. Then the result proved the excellence of the proposed method in terms of accuracy and user satisfaction.

This paper presents an indoor wayfinding system to help the visually impaired finding their way to a given destination in an unfamiliar environment. The main novelty is the use of the user's situation as the basis for designing color codes to explain the environmental information and for developing the wayfinding system to detect and recognize such color codes. Actually, people would require different information according to their situations. Therefore, situation-based color codes are designed, including location-specific codes and guide codes. These color codes are affixed in certain locations to provide information to the visually impaired, and their location and meaning are then recognized using the proposed wayfinding system. Consisting of three steps, the proposed wayfinding system first recognizes the current situation using a vocabulary tree that is built on the shape properties of images taken of various situations. Next, it detects and recognizes the necessary codes according to the current situation, based on color and edge information. Finally, it provides the user with environmental information and their path through an auditory interface. To assess the validity of the proposed wayfinding system, we have conducted field test with four visually impaired, then the results showed that they can find the optimal path in real-time with an accuracy of 95%.

This paper presents scratch restoration method that can deal with scratches of various lengths and widths in old film. The proposed method consists of detection and reconstruction. The detection is performed using texture and shape properties of the scratches: first, each pixel is classified as scratches and non-scratches using a neural network (NN)-based texture classifier, and then some false alarms are removed by shape filtering. Thereafter, the detected region is reconstructed. Here, the reconstruction is formulated as energy minimization problem, thus genetic algorithm is used as optimization algorithm. The experimental result with well-known old films showed the effectiveness of the proposed method.

Many people want to feel safe and secure by monitoring events and changes surrounding them that affect their quality of life. This concern prompted the development of the environment monitoring system. Among the many research groups, sensors and sensor network are their choice of tool for environment monitoring system. Data management is the most challenging issue when we are using sensor networks. To compose sensor network environment need various kind of sensors which does not have standard way to describe sensors and sensor data. In this paper, we will discuss the data management system for the distributed real-time emissions air pollutants monitoring system (DREAM). In this system, we will discuss sensor data modeling and usage of those data.

We demonstrate a novel assistive device which can help the visually impaired or blind people to gain more safe mobility, which is called as "EYECane". The EYECane is the white-cane with embedding a camera and a computer. It automatically detects obstacles and recommends some avoidable paths to the user through acoustic interface. For this, it is performed by three steps: Firstly, it extracts obstacles from image streaming using online background estimation, thereafter generates the occupancy grid map, which is given to neural network. Finally, the system notifies a user of an paths recommended by machine learning. To assess the effectiveness of the proposed EYECane, it was tested with 5 users and the results show that it can support more safe navigation, and diminish the practice and efforts to be adept in using the white cane.

A film restoration has been received considerable attention by many researchers, to support multimedia service of high quality. So far many techniques have been developed, however, such techniques do not permit the reconstruction of all kinds of degradation, because they have been developed based on their own specific environments and assumptions. This paper represents automatic restoration method for various type of degradation region. For this, we develop a stochastic method in MRF-MAP (Markov random field - maximum a posteriori) framework, where the restoration problem is formulated as the minimization problem of the posteriori energy function. Then, to minimize the energy function, we use distributed genetic algorithms (DGAs) that effectively deal with combinatorial problems. To assess the validity of the proposed method, it was tested on natural old films and artificially degraded films, and the results were compared with other methods. Then, the results show that the proposed method is superior to other methods.

between the user and the wheelchair. To facilitate a wide variety of user abilities, the proposed system uses faceinclination and mouth-shape information as user's intention, where the direction of an IW is determined by the inclination of the user's face, while proceeding and stopping are determined by the shape of the user's mouth. This mechanism requires minimal motion, thereby making the system more comfortable and adaptable for the severely disabled. Furthermore, to fully guarantee user's safety, the 10 range-sensors are used to detect obstacles in environment and avoid them. To assess the effectiveness of the proposed IW, it was tested with 34 users and the results show that it can provide a user unable to drive a standard joystick with friendly and convenient system

Community Web sites on specific topics are very popular on the Web. Some active Web communities are so huge and diverse that it becomes a challenging issue to efficiently mine meaningful knowledge from the Web communities. In this paper, we develop schemes to discover and browse power users by their activities in online communities. The novelties of this work are two-fold. 1) We define new features to describe user's social activities: statistical features to summarize userspsila activities and relationship-based features to describe interactions between individual users. And, through extensive user study and experiments to compare the performances of the ranking models based on various features, it is shown that the cross reference (CR) feature plays an unique and effective role in discovering power users in post-dominant online communities. 2) Thereafter, we develop a novel interface for effective exploration of power users based on the CR rank. Two schemes are proposed to incrementally navigate a large number of candidate power users with higher CR values: threshold-based navigation and traversal-based one. Experimental results shows that the proposed CR rank can be used for effective browsing of power users: about 70% precision is maintained while retrieving all the power users, which means that we can discover all the power users with relatively small number of false alarms.

To support the content-based functionalities in the new video coding standard MPEG-4, each frame of a video sequence must first be extracted into video object planes, each of which represents a meaningful moving object. However, extraction and tracking of a video sequence into video object planes remains a difficult and unresolved problem. Accordingly, this paper presents an object extraction and tracking method based on genetic algorithms that can automatically extract and track moving objects. Each frame is spatially segmented by chromosomes that evolve using distributed genetic algorithms. Thereafter, the spatial segmentation result is combined with a change detection mask, thereby creating the video object planes. To eliminate any redundant computation and maintain the temporal continuity of the same objects between two consecutive frames, the chromosomes are started from the spatial segmentation result of the previous frame, and then only unstable chromosomes corresponding to the moving object parts are evolved. Experiment results demonstrate the effectiveness of the proposed method.

An unsupervised method is presented for segmenting video sequences degraded by noise. Each frame in a sequence is modeled using a Markov random field (MRF), and the energy function of each MRF is minimized by chromosomes that evolve using distributed genetic algorithms. To improve the computational efficiency, only unstable chromosomes corresponding to moving object parts are evolved. Experimental results show the effectiveness of the proposed method.

The retrieval of video clips from multimedia databases has been increasingly spotlighted. Texts in videos include useful information for automatic annotation or indexing. Text location is the first step for recognizing the textual information. This paper proposes a neural network-based text location method for news video indexing. Text can be characterized by texture, location, alignment, and font size. The proposed method classifies text pixels and non-text pixels using a network that operates as a set of texture discrimination filters. We find and locate text regions using histogram analysis after removing errors in the classification results. Experimental results show that the proposed method is effective at locating texts.