School of Engineering Department of Electronic and Computer Engineering 132 Department of Electronic and Computer Engineering Projects in Audio Signal Processing Supervisor: CHAU, Kevin / ECE Student: CHAK, Wai Ho / COMP Course: UROP1100, Spring This report aims to investigate the psychoacoustic effects of audio masking and its upward spread due to hearing loss. The ultimate purpose of this report is to incorporate audio masking into a hearing loss simulator previously developed. The prerequisite for this research includes a knowledge of the short-time Fourier transform (STFT), spectral leakage, windowing, audio masking, and the Bark scale. The results show that for normal hearing, the present model for audio masking is sufficiently good that one cannot notice a drop in audio quality between the original and masked sound, music, and speech audios despite the drastic difference between their frequency content. Projects in Audio Signal Processing Supervisor: CHAU, Kevin / ECE Student: HE, Weike / ELEC Course: UROP1000, Summer This project explores the possibility of accurate solo piano transcription without machine learning. With onset detection and short-time Fourier transform, frequency peaks found in an audio file can be identified and mapped to fundamental frequencies of piano keys. This information will be used to construct a piano sheet of the audio file. This project further evaluates the relative performance of this program by comparing its output with piano sheets available on the internet or generated by open-source audio-to-midi software. Results of performance evaluation demonstrate the limitation of Fourier transform and the potential of machine learning models in automatic piano transcription. Robust and Generalized Methods for Medical Image Analysis Supervisor: LI, Xiaomeng / ECE Student: LIU, Yunfei / CPEG MA, Wanqin / ELEC MU, Xihe / CPEG Course: UROP3100, Fall UROP4100, Fall UROP1100, Fall For the UROP program in Fall 2022, we study as a group of three to learn more details about Medical Computer Vision. LIU Yunfei’s part mainly talks about the modifications he made in the paper TINC: Temporally Informed Non-Contrastive Learning for Disease Progression Modeling in Retinal OCT Volumes. MU Xihe focused on active learning and learned basics about traditional segmentation and also participated in some Kaggle tests to familiarize the usage of pytorch library. MA Wanqin’s main task is to read papers and reproduce the experiments related to semi-supervised learning and looking for suitable dataset for research related to class-imbalance. This report will provide information about academic papers and the details of experiments we did in summer, in which LIU contributes for Part 2, MU contributes to Part 3, MA contributes to Part 4.
RkJQdWJsaXNoZXIy NDk5Njg=