Summary
Yihua Zhou is a research scientist specializing in speech recognition with eight years of experience building and deploying edge-oriented keyword spotting, sound event detection, and speaker identification models. Based in Suzhou, he has delivered 20–50% relative improvements in SER for on-device KWS across 20+ real-world testsets while keeping false alarm rates steady and automating data-processing and training with a bespoke toolkit. His background spans NLP model training and production deployment (BERT fine-tuning, contrastive learning, TF-Serving/Triton/FastAPI) and earlier roles in quant analytics and web scraping, giving him a rare mix of signal-processing, ML engineering, and production systems expertise. He frequently researches novel architectures, training strategies, and augmentation tricks and partners closely with SDE and signal-processing teams to port models to chips. Outside his job title, his profile hints at a pragmatic, results-driven approach and a developer mindset—he maintains a personal site and a candid GitHub presence that reflects hands-on experimentation.
8 years of coding experience
3 years of employment as a software developer
Master, Business Analytics, 1st Class Honors Degree, Master, Business Analytics, 1st Class Honors Degree at Xi'an Jiaotong-Liverpool University
Master, Business Analytics, 1st Class Honors Degree, Master, Business Analytics, 1st Class Honors Degree at Liverpool University
English, Japanese