Adela Barbulescu
Researcher and entrepreneur, protecting faces, voices and creative work from unauthorised use by AI.
Bio
Adela Barbulescu is a researcher and entrepreneur working on the protection of identity and content in the age of generative AI. At Inria Startup Studio, she leads research on protecting faces, voices and creative work against unauthorised use for AI model training and generation. Her interests include adversarial machine learning and the robustness of these protections to adaptive attacks.
Her background is in generative modelling of human faces and voices. Her PhD at Inria and GIPSA-lab, awarded by Université Grenoble Alpes, modelled how speech, facial expression and head motion jointly convey emotions and social attitudes, in order to generate expressive virtual actors. She then co-founded Emoface and led it for five years as CEO and scientific director, transferring this research into a digital therapy tool in which expressive 3D avatars support emotion recognition training for autistic children.
She holds an MSc in Vision, Graphics and Interactive Systems from Aalborg University and a BSc in Computer Science from Politehnica University of Bucharest. She is based in Singapore.
Now
Research Engineer and Project Manager, Inria Startup Studio, Grenoble, since Feb 2026.
Protecting images and voices from non-authorised cloning and generative AI model training.
Beepfake: a protection for images and voices against non-consensual AI cloning, and a benchmark of such protections. With Inria and the National University of Singapore (NUS).
Recent writing for a general audience: La violence d’être cloné : et si le deepfake n’était pas qu’un mensonge, mais aussi une atteinte à notre corps numérique ? (The Conversation France, 2026)
Positions
- Feb 2026 – present Research Engineer and Project Manager Inria Startup Studio, Grenoble Protecting images and voices from non-authorised cloning and generative AI model training. Current project: Beepfake.
- Apr 2025 – Jan 2026 Postdoctoral Researcher GIPSA-lab (CNRS), Grenoble Pipeline for audio-visual text-to-speech animations transferable to social robots.
- Jan 2019 – Dec 2024 Co-founder & CEO, Scientific Director Emoface, Grenoble Digital therapy tool in which expressive 3D avatars support emotion recognition training for autistic children.
- Apr 2017 – Oct 2018 Research and Development Engineer GIPSA-lab (CNRS), Grenoble Design, implementation and user testing of the Emoface game prototype in Unity, adapted for individuals with autism.
- Dec 2015 – Nov 2016 Postdoctoral Researcher Inria Rhône-Alpes, Montbonnot Algorithms for generating 3D animations from children’s make-believe games.
- Dec 2012 – Nov 2015 PhD Researcher Université Grenoble Alpes, Grenoble Generating expressive audio-visual speech for virtual actors with emotional context.
- Feb 2012 – Jul 2012 Research Assistant Computer Vision Center, Barcelona Algorithms for estimating 3D human poses from video.
Emoface, 2019–2024
Co-founded and led as CEO and scientific director. The application reached over 20,000 downloads, and the company secured over 800 K€ in R&D grants, with partners including Grenoble Alpes University, Lillab, Multiple (Autism&Tech), MIAI. Watch the Emoface video (YouTube, opens on their site).
Education
- 2012–2015 PhD in Mathematics and Computer Science Université Grenoble Alpes, Grenoble, France Thesis: Generation of audio-visual prosody for expressive virtual actors.
- 2010–2012 MSc in Vision, Graphics and Interactive Systems Aalborg University, Aalborg, Denmark With the Computer Vision Center, Universitat Autònoma de Barcelona, Spain.
- 2006–2010 BSc in Computer Science Politehnica University of Bucharest, Bucharest, Romania
Awards
- France 2030 “Pionniers de l’IA”
- WomenTechEU Prize
- BPI i-Lab National Innovation Competition
- Jacques Chirac Foundation Innovation Prize
- HandiTech Trophy
- Impact Hub Innovation Prize
Publications
Complete record on HAL and Google Scholar.
Writing for a general audience
- La violence d’être cloné : et si le deepfake n’était pas qu’un mensonge, mais aussi une atteinte à notre corps numérique ? Barbulescu, A. (2026). The Conversation France.
Journal articles
- A Generative Audio-Visual Prosodic Model for Virtual Actors Barbulescu, A., Ronfard, R., Bailly, G. (2017). IEEE Computer Graphics and Applications, 37(6), 40–51. Open access on HAL
- Which prosodic features contribute to the recognition of dramatic attitudes? Barbulescu, A., Ronfard, R., Bailly, G. (2017). Speech Communication, 95, 78–86. Open access on HAL
Conference papers
- A system for creating virtual reality content from make-believe games Barbulescu, A., Garcia, M., Bégault, A., et al. (2017). IEEE Virtual Reality, 207–208.
- Figurines, a multimodal framework for tangible storytelling Portaz, M., Garcia, M., Barbulescu, A., et al. (2017). WOCCI 2017, 52–57.
- Making Movies from Make-Believe Games Barbulescu, A., Garcia, M., Vaufreydaz, D., Cani, M.-P., Ronfard, R. (2017). WICED 2017, 51–54.
- Characterization of Audiovisual Dramatic Attitudes Barbulescu, A., Ronfard, R., Bailly, G. (2016). Interspeech 2016, 585–589.
- Audiovisual generation of social attitudes from neutral stimuli Barbulescu, A., Bailly, G., Ronfard, R., Pouget, M. (2015). FAAVSP 2015, 34–39.
- Beyond basic emotions: expressive virtual actors with social attitudes Barbulescu, A., Ronfard, R., Bailly, G., Gagneré, G., Çakmak, H. (2014). MIG 2014, 39–47.
- Audio-visual speaker conversion using prosody features Barbulescu, A., Hueber, T., Bailly, G., Ronfard, R. (2013). AVSP 2013, 11–16.
- Reactive Statistical Mapping: Towards the Sketching of Performative Control with Data D’Alessandro, N., Tilmanne, J., Astrinaki, M., Hueber, T., Dall, R., Ravet, T., Moinet, A., Cakmak, H., Babacan, O., Barbulescu, A., et al. (2013). eNTERFACE 2013, 9th International Summer Workshop on Multimodal Interfaces. Springer, IFIP AICT 425, 20–49.
- 3D human pose estimation using 2D body part detectors Barbulescu, A., Gong, W., Gonzàlez, J., Moeslund, T. B., Roca, F. X. (2012). ICPR 2012, 2484–2487.
Thesis
- Generation of audio-visual prosody for expressive virtual actors Barbulescu, A. (2015). PhD thesis, Université Grenoble Alpes.
Contact
- Email barbulescu.adela@gmail.com
- CV Adela_Barbulescu_CV.pdf
- LinkedIn linkedin.com/in/adelabarbulescu
- Scholar Google Scholar profile
- HAL Publications on HAL