Portrait of David Combei
PhD Student · Machine Learning

David Combei

I work on explainable audio deepfake detection — building detectors that hold up on real-world audio and can show why a recording is flagged as fake.

Artificial Intelligence Research Institute — AIRi@UTCN Technical University of Cluj-Napoca, Romania
Google Scholar LinkedIn Hugging Face Email

About

Making audio deepfake detectors trustworthy

I am a PhD student in Machine Learning at the Artificial Intelligence Research Institute (AIRi@UTCN), Technical University of Cluj-Napoca. My research focuses on explainable audio deepfake detection: understanding what speech anti-spoofing models actually learn, and making their decisions interpretable to the people who rely on them.

I contribute to several research projects, and my work aims to bridge the gap between robust ML models and real-world audio challenges — from in-the-wild deepfakes to source attribution of synthetic speech.

I am working under the supervision of Adriana Stan at the Technical University of Cluj-Napoca.

Research

Collaborations & projects

AIRi@UTCN

Artificial Intelligence Research Institute, Technical University of Cluj-Napoca

AI4TRUST

Horizon Europe project against online disinformation

Solutions

Research project at the Technical University of Cluj-Napoca

Bitdefender

Industry research collaboration

Publications

Selected papers

  1. 2025

    On the Contribution of Lexical Features to Speech Emotion Recognition

    David Combei

    Proc. of the 13th Conference on Speech Technology and Human-Computer Dialogue (SpeD)

    PDF
  2. 2025

    Unmasking real-world audio deepfakes: A data centric approach

    David Combei, Adriana Stan, Dan Oneață, Nicolas Müller, Horia Cucu

    Interspeech 2025

    PDF
  3. 2025

    TADA: Training-free Attribution and Out-of-Domain Detection of Audio Deepfakes

    Adriana Stan, David Combei, Dan Oneață, Horia Cucu

    Interspeech 2025

    PDF
  4. 2024

    WavLM model ensemble for audio deepfake detection

    David Combei, Adriana Stan, Dan Oneață, Horia Cucu

    ASVspoof 2024 Workshop (Automatic Speaker Verification and Spoofing Countermeasures), pp. 170–175

    PDF

Events

Events & presentations

  1. 2025

    Eastern European Machine Learning Summer School (EEML)

    Selected to attend the highly competitive EEML summer school in Sarajevo.

  2. 2025

    Interspeech 2025 — Rotterdam, the Netherlands

    Presented “Unmasking real-world audio deepfakes: A data centric approach” and “TADA: Training-free Attribution and Out-of-Domain Detection of Audio Deepfakes”.

  3. 2024

    Romanian AI Days 2024

    One of the students selected to present their BSc thesis, “Text and Speech Emotion Recognition using pre-trained self-supervised models”.