I am a PhD candidate in the Centre for Speech Technology Research (CSTR). I am supervised by Simon King (primary), Steve Renals, and Junichi Yamagishi. My work explores deep learning approaches to model and control various factors of the speech signal including self-supervised learning and disentanglement. I am also interested in ethical issues surrounding speech technology, including: voice data privacy, voice spoofing/anti-spoofing, and speech technology security.
From February 2020 to February 2021, I was a visiting student in the Yamagishi Lab at the National Institute of Informatics (NII) in Tokyo, Japan. During that time, I collaborated with other PhD students and post-docs to develop techniques that disentangle information in the speech signal for use in speech synthesis.
In 2019, I led a team in the Automatic Speaker Verification ASVspoof Challenge 2019. We submitted two system entries, one for each of the countermeasures tasks: 1) logical access task (text-to-speech, voice conversion, and copy-synthesis), and 2) physical access task (speech intercept and replay under various conditions). In July 2018, I worked on a team that earned First Place for multi-modal emotion recognition in the 2018 ACL Multimodal Challenge using audio, text, and video data collected 'in the wild'. In December 2017, I participated in the Alan Turing Institute Data Study Group (DSG), and worked on a team developing natural language processing analytics and insights from ISIS extremist propaganda for the UK Cabinet Office Defence and Security.
Before starting my PhD, I spent 5 years on the technical staff at MIT Lincoln Laboratory, in the Human Language Technology Group (now the AI Technology and Systems Group). I had the pleasure to work on a variety of research projects which involved rapid prototyping solutions for speech and text data. Prior to joining MIT, I was a visiting scholar at Institute for Infocomm Research (I2R) in Singapore where I worked on simultaneous translation. I earned my Bachelors degree in Applied Linguistics at Portland State University ('09), and my Masters degree in Computational Linguistics at Georgetown University ('12). I have held numerous STEM internships in industry fields such as physics, engineering, and computer science - as well as additional coursework.
My PhD is funded by a generous studentship that is administered by the UoE School of Informatics Centre for Doctoral Training (CDT) in Data Science, and sponsored by the UK Engineering and Physical Sciences Research Council (EPSRC)
Selected Publications
2021
Jennifer Williams, Junichi Yamagishi, Paul-Gauthier Noé, Cassia Valentini-Botinhao, Jean-François Bonastre. "Revisiting Speech Content Privacy". ISCA Symposium on Security and Privacy in Speech Communication (SPSC) 2021. [PDF]
Jennifer Williams, Jason Fong, Erica Cooper, Junichi Yamagishi. "Exploring Disentanglement with Multilingual and Monolingual VQ-VAE". ISCA Speech Synthesis Workshop 2021 (SSW11). [PDF]
Jennifer Williams, Yi Zhao, Erica Cooper, Junichi Yamagishi. "Learning Disentangled Phone and Speaker Representations in a Semi-Supervised VQ-VAE Paradigm". IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2021. [PDF]
2020
Jennifer Williams. "End-to-End Signal Factorization for Speech: Identity, Content, and Style". International Joint Conference on Artificial Intelligence (IJCAI) 2020, Doctoral Consortium. [PDF]
Jennifer Williams, Joanna Rownicka, Pilar Oplustil-Gallegos, and Simon King. "Comparison of Speech Representations for Automatic Quality Estimation in Multi-Speaker Text-to-Speech Synthesis". Speaker Odyssey 2020. [PDF]
2019
Jennifer Williams and Simon King. "Disentangling Style Factors from Speaker Representations". INTERSPEECH 2019. [PDF]
Jennifer Williams and Joanna Rownicka. "Speech Replay Detection with x-Vector Attack Embeddings and Spectral Features". INTERSPEECH 2019.[PDF]
2018
Jennifer Williams, Steven Kleinegesse, Ramona Comanescu, and Oana Radu. "Recognizing Emotions in Video Using Multimodal DNN Feature Fusion". Workshop on Computational Modelling of Human Multimodal Language, ACL 2018. [PDF] ** First Place
Jennifer Williams, Ramona Comanescu, Oana Radu, and Leimin Tian. "DNN Multimodal Fusion Techniques for Predicting Video Sentiment". Workshop on Computational Modelling of Human Multimodal Language, ACL 2018. [PDF]
2017
Jennifer Williams and Charlie Dagli. "Twitter Language Identification of Similar Languages and Dialects Without Ground Truth". Fourth Workshop on NLP for Similar Languages, Varieties and Dialects (VarDial), EACL 2017. [PDF]
Recent PhD Experience
Tutor for Computer Programming for Speech and Language Processing
Research co-supervisor (with Catherine Lai) for MSc dissertation
Topic: Multilingual Representation Learning for Speech Code-Switching and Voice Conversion
Guest Lecture: "Introduction to Ethics for NLP Student Researchers" (for NLP CDT course, Frank Keller)
Invited talk (University of Montreal-IVADO): "Recent Ethical Issues for AI/NLP"
Lab Demonstrator for Computer Programming for Speech and Language Processing
Marker for Accelerated Natural Language Processing
Marker for Machine Learning Practical
Marker for Introductory Applied Machine Learning
Marker for Machine Learning Practical
Research co-supervisor (with Moez Ajili) for undergraduate thesis (SUP'COM, Tunisia)
Topic: Adapting extractive/abstractive text summarization for speech meetings
Marker for Machine Learning Practical
Guest lecture: "Deep Fakes In a Time of Voice-Enabled Consumer Electronics" (for Data and Society course, James Stewart)
Guest lecture/workshop: "Ethics Case Studies" (for NLP CDT course, Adam Lopez)
Research co-supervisor (with Simon King) for visiting post-graduate student (2 month project)
Topic: Improving speaker recognition and factorization from x-vectors
Tutor for Introduction to Applied Machine Learning
Lab Demonstrator and Marker for Computer Programming for Speech and Language
Tutor and Marker for Accelerated Natural Language Processing
Marker for Machine Learning Practical
Research co-supervisor (with Catherine Lai) for MSc dissertation on representation learning
Topic: Emotion/sentiment recognition in multimodal data
Our student earned the Highly Commended Dissertation Prize from the School of Informatics
Tutor and Marker for Machine Learning Practical
Tutor and Marker for Informatics Project Proposal
Lab Demonstrator for Data Mining and Exploration
Lab Demonstrator and Marker for Foundations of Natural Language Processing
Marker for Introduction to Applied Machine Learning
TA for Introduction to Research in Data Science
Tutor, Lab Demonstrator and Marker for Accelerated Natural Language Processing
Tutor for Cognitive Science (undergraduate course)
TA and Lab Demonstrator for Extreme Computing
[In the UK system: Tutor = lead groups of 10-20 students through sample problems and research; Lab Demonstrator = nurture student curiosity about computer programming in the lab; Marker = grade student coursework, project reports, and/or exams; TA = develop course material and hold office hours]
One of my first fascinations with sound came from a summer job that I held as an undergraduate student. I had spent a summer presenting physics experiments to the public at the Oregon Museum of Science and Industry (Portland, Oregon). I had the pleasure of explaining a variety of interesting concepts - one of them was nodal patterns made by sound waves as they travelled through a medium known as a Chladni Plate. Different sound frequencies will produce different nodal patterns, with higher frequency sounds producing more complex patterns.
I also enjoy thinking in the space of philosophy of the mind, including The Extended Mind. My earliest endeavors in linguistics, machine learning, and artificial intelligence were inspired by the work of Andy Clark (Edinburgh) and Hilary Putnam (Harvard) - each in different ways.
In my free time I work on an interesting problem in number theory.
Accessibility Statement
This website is designed to be compliant with screen readers and other accessibility tools. All materials on this website can be provided in alternate accessible formats upon request (example: large print). Please email j.williams@ed.ac.uk to make this request. More details about the accessibility of this website, please click: accessibility.
