MS ECE @ USC, Fall 2026 | Founder, Mantle Sound
Gewenxin Yu
Acoustics, Sound, and Speech Processing
about
A researcher based in Jiangsu, China, I work at the intersection of audio signal processing, bioacoustics, and machine learning. I founded Mantle Sound in 2025 and hold an MSc from Queen Mary University of London (QMUL). In Fall 2026, I will join the University of Southern California (USC) to pursue an MS in Electrical and Computer Engineering, with an eye toward doctoral research in the field.
My work spans from developing neural audio codecs for geophone recordings to creating environmental field recordings; since 2023 I have conducted expeditions across China, the UK, France, and Switzerland, most recently documenting the frozen high-altitude soundscape of Yema Haizi (~4,100 m) on the eastern Tibetan Plateau.
research
Captured ecological soundscapes and built annotation workflow for soundscape corpus. Trained EnCodec-based neural audio codec with Transformer for ultra-low bitrate compression.
Developed two-stage anonymisation system combining McAdams coefficient DSP and VAE-GAN timbre transfer. Trained on Flickr 8k Audio Caption Corpus. Integrated into ZEBRA framework with improved EER and voice distinctiveness metrics.
(Master's thesis supervised by Dr. Charalampos Saitis, Communication Acoustics Lab at QMUL)
sound works
Field recording document from a seasonally frozen lake on the Tibetan Plateau
Interactive sound installation that sonifies Moka Pot movements using IanniX, Arduino, and Max/MSP
code
Built autoencoder-based denoising pipeline for bird song recordings from Warblrb10k dataset. Evaluated using SDR/SIR/SAR metrics. Integrated with CNN classifier for improved bird presence classification.
talks
memberships
- Society for Music Perception and Cognition (SMPC)
- British Wittgenstein Society (BWS)
- The Association Les Amis de Xenakis
- International Playing-Card Society (IPCS)
- Free Software Foundation (FSF)