Search

Humans communicate and interact through natural languages, such as American English (AE), Taiwanese, Italian, and numerous variants of Spanish. Through automatic speech analysis and recognition technologies, human-machine interaction systems (HMISs) can be used for language learning in query systems, smart devices, and healthcare applications, emphasizing the need to enhance user interaction across different sectors. Because people differ in their basic attributes (e.g., gender, age group, and spoken dialect), an HMIS must be able to identify the speaker’s gender, age group, and regional dialect on the basis of their speech signals. To achieve automatic speech recognition, we analyzed and distinguished feature patterns using a feature extraction method and identified gender and region using a convolutional neural network (CNN)-based classifier. Mel-frequency cepstral coefficients were used to extract Mel-scale frequencies (MSF) from dialect-sentence speech signals for conversion into specific feature patterns. Subsequently, a one-dimensional CNN-based classifier was used to identify these features patterns by gender and regional dialect. The proposed speech classifier was rigorously trained, tested, and validated using dialect-sentence speech corpora from AE, Italian (IT), and Spanish (SP) acoustic-phonetic continuous speech database. The experimental results indicate that the proposed model with MSF features can perform accurate gender and region recognition. The classifier was evaluated in metrics of precision (%), recall (%), F1 score, and accuracy (%).

Obstructive sleep apnea (OSA) has been a common sleep disorder for years, and polysomnography (PSG) remains the gold standard for diagnosing OSA. Nevertheless, PSG is a time and money consuming test, and patients have to wait long for arranging a PSG test in a hospital. In light of this, portable and wearable tools for OSA classification have been developed recently as a low‐cost and easy‐to‐use screening method before undergoing PSG. Using unsegmented electrocardiogram (ECG) signals, a deep neural network (DNN)‐based model is developed here to categorize OSA severity with the following features. First, the model takes unsegmented ECG signals recorded overnight as input, and then generates a four‐level scale as output. Since all the input ECG signals are unsegmented, the tremendous amount of effort spent on signal annotation can be fully saved. Second, the largest amount of data is used to test the model and consequently provide a high generalization ability, as compared with others in the literature. The overall outperformance of this work is highlighted at the end of this article, and this work is validated as an easy‐to‐use and effective screening tool for OSA accordingly.

Globally, the incidence rate for breast cancer ranks first. Treatment for early-stage breast cancer is highly cost effective. Five-year survival rate for stage 0–2 breast cancer exceeds 90%. Screening mammography has been acknowledged as the most reliable way to diagnose breast cancer at an early stage. Taiwan government has been urging women without any symptoms, aged between 45 and 69, to have a screening mammogram bi-yearly. This brings about a large workload for radiologists. In light of this, this paper presents a deep neural network (DNN)-based model as an efficient and reliable tool to assist radiologists with mammographic interpretation. For the first time in the literature, mammograms are completely classified into BI-RADS categories 0, 1, 2, 3, 4A, 4B, 4C and 5. The proposed model was trained using block-based images segmented from a mammogram dataset of our own. A block-based image was applied to the model as an input, and a BI-RADS category was predicted as an output. At the end of this paper, the outperformance of this work is demonstrated by an overall accuracy of 94.22%, an average sensitivity of 95.31%, an average specificity of 99.15% and an area under curve (AUC) of 0.9723. When applied to breast cancer screening for Asian women who are more likely to have dense breasts, this model is expected to give a higher accuracy than others in the literature, since it was trained using mammograms taken from Taiwanese women.

A variety of feature extraction and classification approaches have been proposed using electrocardiogram (ECG) and ECG-derived signals for improving the performance of detecting apnea events and diagnosing patients with obstructive sleep apnea (OSA). The purpose of this study is to further evaluate whether the reduction of lower frequency P and T waves can increase the accuracy of the detection of apnea events. This study proposed filter bank decomposition to decompose the ECG signal into 15 subband signals, and a one-dimensional (1D) convolutional neural network (CNN) model independently cooperating with each subband to extract and classify the features of the given subband signal. One-minute ECG signals obtained from the MIT PhysioNet Apnea-ECG database were used to train the CNN models and test the accuracy of detecting apnea events for different subbands. The results show that the use of the newly selected subject-independent datasets can avoid the overestimation of the accuracy of the apnea event detection and can test the difference in the accuracy of different subbands. The frequency band of 31.25–37.5 Hz can achieve 100% per-recording accuracy with 85.8% per-minute accuracy using the newly selected subject-independent datasets and is recommended as a promising subband of ECG signals that can cooperate with the proposed 1D CNN model for the diagnosis of OSA.

Many works in recent years have been focused on developing a portable and less expensive system for diagnosing patients with obstructive sleep apnea (OSA), instead of using the inconvenient and expensive polysomnography (PSG). This study proposes a sleep apnea detection system based on a one-dimensional (1D) deep convolutional neural network (CNN) model using the single-lead 1D electrocardiogram (ECG) signals. The proposed CNN model consists of 10 identical CNN-based feature extraction layers, a flattened layer, 4 identical classification layers mainly composed of fully connected networks, and a softmax classification layer. Thirty-five released and thirty-five withheld ECG recordings from the MIT PhysioNet Apnea-ECG Database were applied to train the proposed CNN model and validate its accuracy for the detection of the apnea events. The results show that the proposed model achieves 87.9% accuracy, 92.0% specificity, and 81.1% sensitivity for per-minute apnea detection, and 97.1% accuracy, 100% specificity, and 95.7% sensitivity for per-recording classification. The proposed model improves the accuracy of sleep apnea detection in comparison with several feature-engineering-based and feature-learning-based approaches.

Polyacrylonitrile-based activated carbon fibers (ACFs), modified using potassium hydroxide (KOH) or tetraethylenepentamine (TEPA), were investigated for carbon dioxide (CO2) adsorption, which is one of the promising alleviation approaches for global warming. The CO2 adsorption isotherms were measured, and the values of isosteric heat of adsorption were calculated. The results showed that the KOH-modified ACFs exhibited a great deal of pore volume, and a specific surface area of 1565 m2/g was obtained. KOH activation made nitrogen atoms easily able to escape from the surface of ACFs. On the other hand, the surface area and pore volume of ACFs modified with TEPA were significantly reduced, which can be attributed to the closing or blocking of micropores by the N-groups. The CO2 adsorption on the ACF samples was via exothermic reactions and was a type of physical adsorption, where the CO2 adsorption occurred on heterogeneous surfaces. The CO2 uptakes at 1 atm and 25 °C on KOH-activated ACFs reached 2.74 mmole/g. This study observed that microporosity and surface oxygen functionalities were highly associated with the CO2 uptake, implying the existence of O-C coordination, accompanied with physical adsorption. Well cyclability of the adsorbents for CO2 adsorption was observed, with a performance decay of less than 5% over up to ten adsorption-desorption cycles.

Accurate QRS detection is an important first step for almost all automatic electrocardiogram (ECG) analyzing systems. However, QRS detection is difficult, not only because of the wide variety of ECG waveforms but also because of the interferences caused by various types of noise. This study proposes an improved QRS complex detection algorithm based on a four-level biorthogonal spline wavelet transform. A noise evaluation method is proposed to quantify the noise amount and to select a lower-noise wavelet detail signal instead of removing high-frequency components in the preprocessing stage. The QRS peaks can be detected by the extremum pairs in the selected wavelet detail signal and the proposed decision rules. The results show the high accuracy of the proposed algorithm, which achieves a 0.25% detection error rate, 99.84% sensitivity, and 99.92% positive prediction value, evaluated using the MIT-BIT arrhythmia database. The proposed algorithm improves the accuracy of QRS detection in comparison with several wavelet-based and non-wavelet-based approaches.

Adaptive multi-rate wideband (AMR-WB) speech codecs have been widely used for high speech quality in modern mobile communication systems, e.g., handheld mobile devices. Nevertheless, a major handicap is that a remarkable computational load is required in the vector quantization (VQ) of immittance spectral frequency (ISF) coefficients of an AMR-WB coding. In view of this, a two-stage search algorithm is presented in this paper as an efficient way to reduce the computational complexity of ISF quantization in AMR-WB coding. At stage 1, an input vector is assigned to a search subspace in an efficient manner using the binary search space-structured VQ (BSS-VQ) algorithm, and a codebook search is performed over the subspace at stage 2 using the iterative triangular inequality elimination (ITIE) approach. Through the use of the codeword rejection mechanisms equipped in both stages, the computational load can be remarkably reduced. As compared with the original version of the BSS-VQ algorithm, the upgraded version provides a computational load reduction of up to 51%. Furthermore, this work is expected to satisfy the energy saving requirement when implemented on an AMR-WB codec of mobile devices.

A novel tone detection approach, designated as the multi-frequency detecting (MFD) algorithm, is presented in this work as an alternative to conventional single point detection approaches but it is an efficient way to achieve the aim of further computational load reduction for a dual-tone multi-frequency (DTMF) signal detection. The idea is that an optimal phase search is performed over the frequency band of interest in each tone detection, and then the optimal frequency response of a detector is built accordingly. In this manner, a DTMF detection task is done following one-time detection computation. This proposal demonstrates an overall computational load reduction of 80.49% and 74.06% in comparison with a discrete Fourier transform (DFT) approach and the Goertzel algorithm, respectively. This detection complexity reduction is an advantage and an important issue for applying DTMF detection technique to embedded devices.

The adaptive multi-rate wideband (AMR-WB) speech codec is widely used in modern mobile communication systems for high speech quality in handheld devices. Nonetheless, a major disadvantage is that vector quantization (VQ) of immittance spectral frequency (ISF) coefficients takes a considerable computational load in the AMR-WB coding. Accordingly, a binary search space-structured VQ (BSS-VQ) algorithm is adopted to efficiently reduce the complexity of ISF quantization in AMR-WB. This search algorithm is done through a fast locating technique combined with lookup tables, such that an input vector is efficiently assigned to a subspace where relatively few codeword searches are required to be executed. In terms of overall search performance, this work is experimentally validated as a superior search algorithm relative to a multiple triangular inequality elimination (MTIE), a TIE with dynamic and intersection mechanisms (DI-TIE), and an equal-average equal-variance equal-norm nearest neighbor search (EEENNS) approach. With a full search algorithm as a benchmark for overall search load comparison, this work provides an 87% search load reduction at a threshold of quantization accuracy of 0.96, a figure far beyond 55% in the MTIE, 76% in the EEENNS approach, and 83% in the DI-TIE approach.

An adaptive multi-rate wideband (AMR-WB) code is a speech codec developed on the basis of an algebraic code-excited linear-prediction (ACELP) coding technique, and has a double advantage of low bit rates and high speech quality. This coding technique is widely used in modern mobile communication systems for a high speech quality in handheld devices. However, a major disadvantage is that a vector quantization (VQ) of immittance spectral frequency (ISF) coefficients occupies a significant computational load in the AMR-WB encoder. Hence, this paper presents a triangular inequality elimination (TIE) algorithm combined with a dynamic mechanism and an intersection mechanism, abbreviated as the DI-TIE algorithm, to remarkably improve the complexity of ISF coefficient quantization in the AMR-WB speech codec. Both mechanisms are designed in a way that recursively enhances the performance of the TIE algorithm. At the end of this work, this proposal is experimentally validated as a superior search algorithm relative to a conventional TIE, a multiple TIE (MTIE), and an equal-average equal-variance equal-norm nearest neighbor search (EEENNS) approach. With a full search algorithm as a benchmark for search load comparison, this work provides a search load reduction above 77%, a figure far beyond 36% in the TIE, 49% in the MTIE, and 68% in the EEENNS approach.

With the large availability of protein interaction networks and microarray data supported, to identify the linear paths that have biological significance in search of a potential pathway is a challenge issue. We proposed a color-coding method based on the characteristics of biological network topology and applied heuristic search to speed up color-coding method. In the experiments, we tested our methods by applying to two datasets: yeast and human prostate cancer networks and gene expression data set. The comparisons of our method with other existing methods on known yeast MAPK pathways in terms of precision and recall show that we can find maximum number of the proteins and perform comparably well. On the other hand, our method is more efficient than previous ones and detects the paths of length 10 within 40 seconds using CPU Intel 1.73GHz and 1GB main memory running under windows operating system.

In this paper, we propose a method of classifying the severity of obstructive sleep apnea (OSA) using electrocardiogram (ECG) signals and deep learning. In our previous research, we presented an ECG-based signal segmentation-free model for OSA severity classification. Its key feature is using the unsegmented overnight ECG signal as input and directly predicting the four categories of OSA severity as output. The overall performance of our previous work has been demonstrated to significantly exceed those of most existing studies. On the basis of a preliminary study, a method of improving the accuracy of OSA severity classification is proposed in this paper. Modifications to the model architecture for OSA severity classification were made, and a squeeze-and-excitation network (SENet) was integrated into this work. Finally, our experimental results indicated that the accuracy of the four-category classification of OSA severity in this paper is 57.91%, which is slightly higher than 57.55% achieved in our previous research. [ABSTRACT FROM AUTHOR]

Breast cancer is a prevalent global health concern and the most commonly diagnosed cancer in women. Mammography, a well-established and widely used screening tool, has greatly contributed to early breast cancer detection. However, understanding mammographic breast composition is also crucial for refining the risk assessment of breast cancer beyond identifying lesions. In contrast to previous studies, we adopt an exploratory approach by using the Swin Transformer, a foundation model for image classification, to classify the four-category breast density. Leveraging this foundation, we fine-tune the model with a small set of mammograms for the purpose of making advancements in breast density classification. This study is experimentally validated to achieve an overall accuracy of 74.96% in the four-category breast density classification, which is a comparable performance to recent counterparts. [ABSTRACT FROM AUTHOR]

In this paper, we present a study of the license plate recognition (LPR) system for Taiwanese vehicles using a cascade of You Only Look Once version 4 (YOLOv4) detectors. The LPR system is composed of a vehicle detection model, a license plate (LP) detection model, an LP corner prediction model, and an LPR model. Herein, the pretrained YOLOv4 model was directly applied to vehicle detection. The YOLOv4 framework was adopted in the LP detection and LP recognition models, performing transfer learning on each model. Furthermore, to enhance the accuracy of the LPR system, an LP corner prediction model was developed in this study to predict the four corner positions of an LP to perform a perspective transformation on the plate for alignment purposes. The experimental results show that our LPR system achieves an accuracy of 98.88% when tested on 2049 images of the application-oriented LP dataset, outperforming most LPR systems reported in the literature. [ABSTRACT FROM AUTHOR]

This paper presents a teaching tool for schoolchildren to learn artificial intelligence (AI) technology through which a variety of banknotes can be recognized. This was done by first using a pretrained YOLOv3 object detection model. Secondly, transfer learning was conducted on the pretrained model using 11 collected banknotes, including US$, Euro, Japanese Yen, and NT$. The banknote detection model was experimentally validated to give an average precision (AP) of up to 99.09% if the threshold Intersection over Union (IoU) is not less than 0.8. Once a banknote was successfully recognized, the face value and the country name thereon were displayed, and schoolchildren can access suggested websites, i.e., Wikipedia, Google Maps, and the Bank of Taiwan, to learn more about the exchange rate between currencies and the history and location of the country that issued the banknote. Consequently, schoolchildren can have fun using this tool and acquire a more a global outlook. Moreover, they may be motivated to become AI professionals in the future. [ABSTRACT FROM AUTHOR]