Aug 2023 – Jun 2024 · 10 mos · Singapore · Hybrid

• Supervisor: Ho Thi Nga
• Conducted a systematic investigation into Inverse Text Normalization (ITN), a crucial task in Natural Language Processing (NLP). Identified and analyzed mainstream methodologies used in solving ITN, laying the foundation for further research and innovation.
• Reproduced and conducted experiments with state-of-the-art deep learning models for solving ITN using Python and PyTorch, fairseq libraries.
• Developed a Flask APP (webpage) to encapsulate the models in the backend, accompanied by a user-friendly Web UI. The implementation noticeably enhanced the user experience, streamlining the interaction with the ITN models.
• Dockerized the Flask APP into a container, ensuring seamless deployability on various machines. This containerization contributes to project scalability and ease of deployment across different environments.
• Demonstrated initiative by working independently and overcoming challenges through self-learning, showcasing adaptability and a proactive approach to problem-solving.

Dec 2022 – May 2023 · 5 mos · Singapore · On-site

• Developed a new adversarial attack method against the automatic speaker verification systems protected by spoofing countermeasures by incorporating text-to-speech synthesis and I-FGSM. This work is published on the Interspeech 2023 with the title: DoubleDeceiver: Deceiving the Speaker Verification System Protected by Spoofing Countermeasures and I am the first author.
• https://www.isca-speech.org/archive/pdfs/interspeech_2023/zhang23c_interspeech.pdf
• Exercised critical thinking through conducting systematical experiments to verify the effectiveness of new methods proposed by us over previous public methods and also through proposing improvements to overcome some shortcomings of the current methods.
• Gained in-depth and hands-on experience in Python and PyTorch through reproducing model extraction methods, adversarial attack methods, and refactoring code.