My research focuses on advancing Automatic Speech Recognition (ASR) through the design of parameter-efficient architectures, large-scale synthetic data augmentation, and post-training strategies aimed at improving robustness, adaptability, and fairness across diverse languages, speakers, and domains.
Who I am
I am a Postdoctoral Researcher at INESC‑ID (Lisbon) working on parameter‑efficient Transformers and speech technologies. I focus on ASR/TTS, synthetic data augmentation, and post‑training strategies for robustness and fairness.
Selected work
Peer-reviewed and preprint work across ASR, PEFT, and evaluation. See the dedicated page for the full list.
Topics I'm interested in
Children and Low-ressource ASR. Building large-scale TTS and filtering pipelines to expand training corpora and improve robustness.
Integrating acoustic and linguistic features for pathology detection and ensuring that speech technologies are interpretable, inclusive, and privacy-preserving.
Exploring transfer learning, fine-tuning, and domain adaptation methods to enhance the adaptability and robustness of ASR and large language models (LLMs).
Past and present work
Development of state-of-the-art ASR systems for European Portuguese, as well as CAMÕES a comprehensive benchmark for ASR evaluation.
Collecting children's speech from the wild. This project aims to create a diverse dataset that reflects real-world scenarios for testing purposes.
Exploring Exploring Transformer variant that uses loop mechanism for parameter-efficiency.
Explainers & deep dives
Let’s collaborate
Email: trolland@protonmail.ch · GitHub · LinkedIn · CV (PDF)