School of Engineering Department of Electronic and Computer Engineering 159 Projects in Audio Signal Processing Supervisor: Kevin CHAU / ECE Student: LUI Sin Hang / ELEC Course: UROP 1100, Fall UROP 2100, Spring In recent years, people have created numerous deep-fake technologies, some of which have been exploited by criminals to participate in unlawful acts, posing an increasing threat. The most important one is Phone Scamming. Currently, phone scammers can mimic the voices of the victims’ families or friends to gain their trust and consequently swindle their money more effortlessly. Per the Q4 2024 Global Call Threat Report published by Hiya, a worldwide organization focused on combating spam occurrences, AI-generated impersonation scams are on the rise, with almost 30% of consumers in Canada, the U.S., and the U.K. experiencing deepfake fraud calls. Hence, we aim to explore emotion recognition from audio as well as deepfake audio. Projects in Audio Signal Processing Supervisor: Kevin CHAU / ECE Student: YANG Dongxiao / CPEG Course: UROP 1100, Spring This report presents a digital conducting wand system for real-time control of audio tempo and volume without altering pitch. Using Python with libraries like librosa and pyrubberband, the software component enables tempo adjustments via keystrokes, achieving smooth transitions with low latency. The system processes WAV files, preserving audio quality through time-stretching algorithms. Future work involves integrating a hardware wand with motion sensors to map gestures to tempo and volume changes. Preliminary results demonstrate precise tempo control, with potential applications in music education and live performances. This interdisciplinary approach combines signal processing and human-computer interaction, paving the way for intuitive audio manipulation. Projects in Audio Signal Processing Supervisor: Kevin CHAU / ECE Student: YAU Man Kit Bosco / CPEG Course: UROP 1100, Fall This report explores the development and evolution of musical instrument recognition techniques in Western classical music. First the report discusses fundamental signal analysis concepts, including sampling theory and Fourier transforms, which form the mathematical foundation for these recognition systems. We examine two major approaches: the pattern recognition method and modern deep learning techniques. The pattern recognition method utilizes log-lag correlograms and 31 distinct features to achieve 71.6% accuracy in specific instrument identification, with better results for broader categorizations. The modern approach employs convolutional neural networks (CNNs) with mel-frequency cepstral coefficients (MFCC), achieving up to 93% accuracy across various instruments. The report concludes by examining future challenges, particularly in mixed instrument recognition, and potential commercial applications in music streaming and production software.
RkJQdWJsaXNoZXIy NDk5Njg=