Thuan Van Tran

Dr. Thuan Van Tran

orcid

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
20 April 2020 in IEEE Access
Verified
Journal Article
Open access
TTThuan Van Tran
Thuan Van Tran
WTWei-Ho Tsai
Wei-Ho Tsai
This work investigates how to detect emergency vehicles such as ambulances, fire engines, and police cars based on their siren sounds. Recognizing that car drivers may sometimes be unaware of the siren warnings of the emergency vehicles, especially when in-vehicle audio systems are used, such that they may block or even collide with the emergency vehicles. Therefore, we aim to develop an automatic detection system that determines whether there are siren sounds from emergency vehicles nearby to alert other vehicles’ drivers to pay attention. We propose a convolutional neural network (CNN)-based ensemble model (SirenNet) with two network streams to classify sounds of traffic soundscape to siren sounds, vehicle horns, and noise, in which the first stream (WaveNet) directly processes raw waveform, and the second one (MLNet) works with a combined feature formed by MFCC (Mel-frequency cepstral coefficients) and log-mel spectrogram. Our experiments on a diverse dataset with the SirenNet showed a promising accuracy of 98.24%, also indicating that the raw feature can complement the MFCC and log-mel features in siren sound detection. Besides, the proposed system can work appropriately with the changeable input length; even with short samples of 0.25 seconds, it still yielded a high accuracy of 96.89%. The results of this work not only be able to help drivers reduce car accidents, but also provide a necessary safety function for autopilot systems.

12 November 2021 in IEEE Sensors Journal
Verified
Journal Article
Closed access
TTThuan Van Tran
Thuan Van Tran
WTWei-Ho Tsai
Wei-Ho Tsai
Emergency vehicles (EVs) are permitted to travel at high speed to quickly reach the destination with the aid of audible and visual warning signals, and other road users are required to clear the path for EVs. However, car drivers may sometimes be unaware of nearby EVs, leading to delay in response or even traffic collisions. This work proposes audio-based and vision-based EV detection systems (EVD) that can detect EVs and alert car drivers to respond appropriately. First, we propose a modified YOLO model tailored to the EVD problem, namely YOLO-EVD, and develop a novel image dataset for vision-based EVD (V-EVD). We utilize cross-stage partial connections at the YOLO-EVD’s neck to enhance the detection performance, in which YOLO-EVD achieves 95.5% mean average precision that is better than those of the other single-stage object detectors. Second, we propose WaveResNet, an end-to-end convolutional neural network, for audio-based EVD (A-EVD) based on the classification of siren sound and traffic noise. With raw waveform input of at least one second, the WaveResNet attains high accuracies, at above 98% in normal traffic, and is robust to noise. Both YOLO-EVD and WaveResNet meet the real-time operation requirement. Also, we integrate YOLO-EVD and WaveResNet to develop a prototype of the audio-vision EVD system (AV-EVD) that is a novel approach in the literature of the EVD problem. Our experiments show the promising results of the AV-EVD system as it produces a low misdetection rate of 1.54%. The proposed A-EVD, V-EVD, and AV-EVD systems can be applied to provide safety functions for private cars, self-driving cars, or smart road infrastructure.

3 October 2019 in IEEE Sensors Journal
Verified
Journal Article
Closed access
TTThuan Van Tran
Thuan Van Tran
WTWei-Ho Tsai
Wei-Ho Tsai

22 June 2022 in IEEE Access
Verified
Journal Article
Open access
TTThuan Van Tran
Thuan Van Tran
WTWei-Ho Tsai
Wei-Ho Tsai
In the places of railroad crossings, audible warning signals such as train whistles and railway alarms are utilized to warn the road users of paying attention and giving priority to the approaching train(s). However, road users may sometimes be unaware of warning signals due to various reasons, resulting in inappropriate cooperation or even traffic collision between railway vehicles and non-railway vehicles. This work studies deep learning-based approaches to develop systems for acoustic-based train arrival detection (A-TAD). Firstly, we develop a novel audio dataset of train horns, railway alarms, railway noise, and other urban noises to conduct A-TAD experiments. We then examine the efficiency of handcrafted acoustic features (i.e. MFCC and Mel-spectrogram) in building A-TAD’s audio classifier, the MSNet, which is based on two-dimensional convolutional neural networks (2D-CNN). Next, we propose to apply the attention mechanism and utilize MFCC and spectrogram simultaneously to enhance the classification accuracy, in which the combined use of acoustic features is considered at the input level (with InCom-TADNet), high-level feature level (with FCCom-TADNet), and decision level (with DLCom-TADNet). Our experiments have shown the efficiency of MSNet and attention mechanism as the MSNet trained with the single feature is more performant than the baseline models and applying attention modules results in better accuracies. Also, the combined use of MFCC and spectrogram significantly improve the system’s accuracy and robustness. A-TAD systems can be utilized to extend the safety function of the railway crossing systems, private cars, and self-driving cars, and particularly be useful for hearing-impaired road users.

17 July 2020 in IEEE Access
Verified
Journal Article
Open access
TTThuan Van Tran
Thuan Van Tran
WTWei-Ho Tsai
Wei-Ho Tsai
Although numerous works have studied the problem of automatic speaker identification (SID), there are only few works on the SID for overlapping speech, and none of them consider the case of more than two simultaneous speakers. Recognizing that overlapping speech occurs frequently in real-life scenarios, such as in meetings or debates, this work investigates the methods for overlapping SID (OSID) that can determine identities in the overlapping speech from up to five simultaneous speakers. We propose two deep-learning OSID systems, one is two-stage and the other is single-stage. The two-stage system determines the number of simultaneous speakers firstly, followed by identifying the speaker(s). The single-stage system uses a single classifier to perform OSID directly, which is slightly more computationally efficient than the two-stage system. Our experiments show that the two-stage OSID system achieves better identification accuracy than that of the single-stage system. In addition, both the OSID systems based on one-dimensional convolutional neural networks (1DCNN) perform better than the systems based on multilayer perceptron (MLP) and Gaussian mixture models (GMMs). The proposed 1DCNN-based two-stage OSID system achieves 98.55% OSID accuracy for the clean audio data containing up to five simultaneous speakers. In more challenging experimental conditions involving both background noises and high overlapping energy ratios, the system still attained accuracies of above 90%.

24 May 2023 in IEEE Access
Verified
Journal Article
Open access
TTThuan Van Tran
Thuan Van Tran
YLYih-Lon Lin
Yih-Lon Lin
WTWei-Ho Tsai
Wei-Ho Tsai
This study examines the use of breath sounds intrusively recorded by mobile devices for person identification (PID), which is referred to as mobile-sensed BreathPID. A custom dataset of breath sounds from 21 volunteers is prepared for investigation and analysis. To overcome the problem of spare training data, we incorporate various audio data augmentation (DA) methods with the self-supervised learning (SSL) approach to train the BreathPID’s learning models. SSL-based models for BreathPID are developed in two phases: firstly, solving proposed pretext task(s) without identity information to effectively learn core characteristics of data; then further finetuning the models on the labeled data for the downstream BreathPID task. Several types of pretext or auxiliary tasks are investigated. First, when considering each DA technique, the pretext task is defined as the detection of augmentation levels, for instance, the levels of noise added to original data samples. When utilizing multiple DA techniques, the identification of DA types is defined as the pretext task. In addition, various issues in developing robust BreathPID systems are taken into consideration, including network design, changes in input length, and the ability of noise resistance. From the experimental results, we find that SSL-based BreathPID with the combined use of four DA techniques (i.e., noise addition, speed changing, time shifting, and spectrogram masking) achieves promising results which are higher than those of SSL-based models using single DA technique and those of typical supervised models. Also, the proposed system shows good resistance to noise effects and changes in the input size. Mobile-sensed BreathPID achieves the equivalent or superior results compared to stethoscope-sensed BreathPID (where breath sounds are sensed using specialty stethoscopes). The proposed approach can be applied to the authentication function or health monitoring applications on mobile devices.