An Innovative Methodology Utilizing AI-based Automatic Speech Recognition for Transcribing Dutch Patient-Provider Consultation Recordings

  • Cristian Tejedor-García (Creator)
  • Henk van den Heuvel (Creator)
  • Arjan J. van Hessen (Creator)
  • Sandra van Dulmen (Creator)
  • Toine Pieters (Creator)

Dataset

Description

This paper introduces an innovative methodology developed within the Homo Medicinalis (HoMed) project for adapting automatic speech recognition (ASR) models to handle sensitive audio data domains, specifically focusing on privacy-sensitive patient-provider medical consultations. By utilizing AI and deep learning algorithms, the project successfully addresses GDPR compliance while enhancing ASR technology's performance through domain-specific adaptation. The methodology involves adapting and testing three state-of-the-art ASR models for Dutch using highly sensitive audio-visual recordings, with data preparation emphasizing privacy-compliant methods and secure training environments. Through this approach, the study not only advances ASR technology's capabilities in specialized audio domains but also provides a blueprint for developing publicly shareable, domain-specific ASR engines while safeguarding the confidentiality of sensitive audio information.
Date made available26 Apr 2024
PublisherZenodo

Cite this