Fine-Tuning OpenAI Whisper-Small for Domain-Specific Medical Speech Recognition within a Microservice Architecture

Abstract

We fine-tune Whisper-small (244M parameters) on 8.5 hours of in-domain medical audio and evaluatewith word error rate (WER). Compared to an unadapted Whisper-small baseline, our fine-tuned modelreduces WER from ∼63% to ∼32%. While the relative gain is substantial, this accuracy is not suitablefor unsupervised clinical use; we position the system as a clinician-in-the-loop assistant. We also describedeployment as an on-premise microservice and report latency/throughput considerations.

Authors

  • Alaeddine Moussa Université de la Manouba, École Nationale des Sciences de l’Informatique, Tunisia
  • Noursene Drine Universite de Tunis El Manar, École Nationale des Sciences de l’Informatique, Tunisia

DOI:

https://doi.org/10.31449/inf.v50i6.12075

Downloads

Published

02/21/2026

How to Cite

Moussa, A., & Drine, N. (2026). Fine-Tuning OpenAI Whisper-Small for Domain-Specific Medical Speech Recognition within a Microservice Architecture. Informatica, 50(6). https://doi.org/10.31449/inf.v50i6.12075

Issue

Vol. 50 No. 6 (2026): Online-only issue

Section

Online-only

License

Authors retain copyright in their work. By submitting to and publishing with Informatica, authors grant the publisher (Slovene Society Informatika) the non-exclusive right to publish, reproduce, and distribute the article and to identify itself as the original publisher.

All articles are published under the Creative Commons Attribution license CC BY 3.0. Under this license, others may share and adapt the work for any purpose, provided appropriate credit is given and changes (if any) are indicated.

Authors may deposit and share the submitted version, accepted manuscript, and published version, provided the original publication in Informatica is properly cited.