Themos Stafylakis
Themos Stafylakis
Associate Professor · Machine Learning & Speech
Associate Professor, Dept. of Informatics, AUEB
Director, MSc in AI & Data Science · AUEB
Director, Information Processing Laboratory · AUEB
Conversational AI Architect, Omilia
Affiliated Researcher, Archimedes / Athena Research Center

I work on machine learning for speech and audio — with a focus on speaker recognition, audiovisual speech processing, voice biometrics, and NLP. I received my PhD from NTUA (2011), held postdoctoral positions at ÉTS Montréal and CRIM (2011–2016), and at the University of Nottingham as a Marie Skłodowska-Curie Fellow (2016–2018). I was elected Associate Professor at AUEB in 2023, where I direct the Information Processing Laboratory and the MSc Programme in AI & Data Science.

Google Scholar LinkedIn Email Download CV
01 / Research

Research Interests

My research sits at the intersection of deep learning, speech processing, and biometrics. I am particularly interested in representation learning for audio and speech, with applications ranging from speaker verification to multimodal understanding. Current directions include self-supervised models for speech, parameter-efficient adaptation, and neural diarization.

Speaker Recognition & Biometrics

Speaker embeddings, i-vectors, PLDA, end-to-end verification, voice anti-spoofing, and domain adaptation for speaker identity.

Audiovisual Speech

Lip reading, audiovisual fusion, end-to-end ASR, and multimodal models for robust speech perception.

Self-Supervised Learning

Representation learning from unlabelled speech with HuBERT, Wav2Vec 2.0, WavLM, and downstream adaptation strategies.

NLP & Spoken Language Understanding

Intent classification, task-oriented dialogue, and end-to-end architectures bridging ASR and NLU.

Speech Emotion Recognition

Attentive pooling methods, label smoothing, and cross-corpus generalization for emotion and affect recognition.

Neural Speaker Diarization

End-to-end neural diarization, Perceiver-based attractor models, and overlap-aware segmentation.

02 / Publications

Selected Publications

2026
AAAI
MMAU-Pro: A Challenging and Comprehensive Benchmark for Holistic Evaluation of Audio General Intelligence
S. Kumar, Š. Sedláček, V. Lokegaonkar, F. López, W. Yu, N. Anand, H. Ryu, L. Chen, M. Plicka, M. Hlaváček, T. Stafylakis, et al.
[Paper] [Project page] [Code]
EACL Workshop
LabelBuddy: An Open Source Music and Audio Language Annotation Tagging Tool Using AI Assistancen
I. Prokopiou, I. Sina, A. Kounelis, P. Vikatos, T. Stafylakis
[arXiv]
2025
SigDial
Building Open-Retrieval Conversational Question Answering Systems by Generating Synthetic Data and Decontextualizing User Questions
C. Vlachos, N. Stylianou, A. Fiotaki, S. Methenitis, E. Palogiannidi, T. Stafylakis I. Androutsopoulos
[arXiv]
Interspeech
Synthetic Speech Source Tracing using Metric Learning
D. Koutsianos, S. Zacharopoulos, Y. Panagakis, T. Stafylakis
[arXiv]
ICASSP
CA-MHFA: A context-aware multi-head factorized attentive pooling for SSL-based speaker verification
J. Peng, L. Mošner, L. Zhang, O. Plchot, T. Stafylakis, Lukas Burget, Jan Cernocky
[arXiv]
2024
IEEE/ACM TASLP
DiaPer: End-to-End Neural Diarization With Perceiver-Based Attractors
F. Landini, M. Díez, T. Stafylakis, L. Burget
[Paper]
ACL Findings
Comparing Data Augmentation Methods for End-to-End Task-Oriented Dialog Systems
C. Vlachos, T. Stafylakis, I. Androutsopoulos
[Paper]
Interspeech
Challenging Margin-Based Speaker Embedding Extractors Using the Variational Information Bottleneck
T. Stafylakis, A. Silnova, J. Rohdin, O. Plchot, L. Burget
[Paper]
2023
EMNLP
A Simple Baseline for Knowledge-Based Visual Question Answering
A. Xenos, T. Stafylakis, I. Patras, G. Tzimiropoulos
[Paper]
ICASSP
Speech-Based Emotion Recognition with Self-Supervised Models Using Attentive Channel-Wise Correlations and Label Smoothing
S. Kakouros, T. Stafylakis, L. Mosner, L. Burget
[Paper]
Highly Cited
Interspeech
Combining Residual Networks with LSTMs for Lipreading
T. Stafylakis, G. Tzimiropoulos
[arXiv]
ICASSP
End-to-End Audiovisual Speech Recognition
S. Petridis, T. Stafylakis, P. Ma, F. Cai, G. Tzimiropoulos, M. Pantic
[arXiv]
CVIU
Pushing the Boundaries of Audiovisual Word Recognition Using Residual Networks and LSTMs
T. Stafylakis, M. H. Khan, G. Tzimiropoulos
[arXiv]
View full publication list on Google Scholar →
03 / PhD Students

PhD Students

I am currently supervising PhD students at AUEB. If you are interested in joining the group, please get in touch by email.

PP
Pavlos Poulos
Brain-inspired AI — novel architectures informed by principles of neuroscience and brain function
AUEB · Archimedes/Athena R.C.
DK
Dimitris Koutsianos
Audiovisual deepfakes and biometric verification
AUEB · Archimedes/Athena R.C.
PK
Pieris Kalligeros
Generative modelling for proteins using Protein Language Models (PLMs)
AUEB · Archimedes/Athena R.C.
GP
Giannis Prokopiou
Machine learning and AI for music
AUEB
04 / Talks

Talks & Invited Lectures

Jul 2024
Self-Supervised Learning for Speech: Architectures, Discrete Units & Downstream Adaptation Invited Tutorial
ACAI / ESSAI 2024 — 21st Advanced Course on Artificial Intelligence  ·  Athens, Greece
2025
Advances in Speaker Recognition & Discrete Audio Tokens Invited
Conversational AI Reading Group — Mila / Concordia University, Montréal  ·  ▶ Recording
2024–present
Athens NLP Summer School (AthNLP) Co-organizer
Annual summer school on Natural Language Processing & Machine Learning — Athens, Greece  ·  Editions: 2024, 2025, 2026
05 / Teaching

Teaching

Courses at the Department of Informatics, Athens University of Economics and Business.

Undergraduate
Signals, Systems & Digital Signal Processing BSc
Department of Informatics · AUEB
Undergraduate
Machine Learning BSc
Department of Informatics · AUEB
Postgraduate
Deep Learning MSc
Department of Informatics · AUEB

Previously also taught "Speech Synthesis and Recognition" and "Dialog Systems" at the Interdisciplinary MSc Programme in Language Technology, National and Kapodistrian University of Athens, in collaboration with the Institute for Language and Speech Processing (ILSP / "Athena" R.C.) — 2022–2024.