D.A. Balan (Dragos)

I have started my Ph.D. as part of the Hoogwaardige Spraakherkenning voor al het Nederlands (HOSAN) project, focusing on creating a Dutch foundation speech recognition model that improves upon the state-of-the-art for all types of Dutch speakers and speech styles. A consortium was established for this project, consisting of partners from Radboud University Nijmegen, Sound & Vision, SURF, and Politie, and the first phase was funded for the year 2025.

In the 3rd and 4th year of my Ph.D., I will be working on the MediSpeech project, which focuses on creating a patient-doctor conversation transcriber pipeline that takes the audio of the conversation as input and creates summaries of the session in Electronic Healthcare Records (EHR) format. This is a 3-year international project with partners from 5 countries: Finland, the Netherlands, Portugal, Turkey, and the United Kingdom. On the Dutch level, the collaboration consists of both industry and academia partners, including Radboud University Nijmegen, TNO, HealthTalk and Amsterdam UMC, and the focus use case is for nephrology patients.

I have started working at University of Twente as a junior researcher, in collaboration with Nederlands Instituut voor Beeld & Geluid (Netherlands Institute for Sound & Vision), on the Oral History - Stories at the Museum around Artworks (OH-SMArt) project. OH-SMArt was a long-term initiative to significantly improve the digital research chain around using Oral History and spoken narratives, with research into artworks and museums as a use case. My role within this project was to evaluate state-of-the-art speech recognition models, such as Whisper and XLS-R, and implement the best-performing model (Whisper large-v2) as an automatic transcriber within the infrastructure of the Institute, which can be used as a tool by researchers working with digital media archives to automatically transcribe Oral History interviews or other audiovisual content present in the Institute's digital archive.

My background is in Computer Science & Engineering (Bachelor's degree from TU Delft) and Voice Technology (Master's degree from University of Groningen - Campus Fryslân). My general interests are speech technology-related, focusing on speech recognition for low-resource contexts (languages, groups of speakers, domain-specific terminology, etc.).

Expertise

• Computer Science

  • Speech Recognition
  • Research Infrastructure
  • Speech Recognition System
  • Benchmarking
  • Open Source
  • Voice Activity Detection
  • Infrastructure Provider
  • Telephone

Organisations

My general research interest is Automatic Speech Recognition, with a focus on large-scale processing within Research Infrastructures and for the Dutch language.

Publications

Research profiles

Address

University of Twente

Citadel (building no. 09), room H234
Hallenweg 15
7522 NH Enschede
Netherlands

Navigate to location

Organisations