The growing demand for speech-to-text applications is transforming the way we approach everyday tasks, enabling hands-free dictation, empowering those with visual impairments, and facilitating transcription services for those struggling with auditory comprehension. Researchers are discovering various applications for these instruments, with Bożena Kostek from Gdańsk College of Technology leading an investigation into the potential of STT to revolutionize medical practices. She aims to identify the extent to which clear speech influences speech-to-text (STT) accuracy, ultimately seeking to elevate its utility for healthcare practitioners.
“Automating note-taking for patients’ information is crucial for doctors and radiologists, as it enables them to devote more face-time with patients while facilitating enhanced data collection,” Kostek notes.
While discussing their project, Kostek also highlights the obstacles they encounter in their endeavour.
While STT (Speech-to-Text) fashions often struggle to accurately process medical terminology, especially when it’s written in Polish, this is largely due to the fact that many of these systems have been trained primarily on English language data. Most sources focus on straightforward language, omitting specialized medical terminology. Noisy hospital environments further test the durability of communication, as stressed or distracted healthcare providers often struggle to convey crucial information effectively.
To address these key considerations, a comprehensive audio dataset was developed featuring Polish medical terminology uttered by healthcare professionals, including doctors and specialists from fields such as cardiology and pulmonology. The dataset was analyzed using an automated speech recognition model, a technology capable of converting spoken language into written text, for transcription purposes. Metrics analogous to Phrase Error Rate and Character Error Rate have been employed to evaluate the quality of speech recognition. This evaluation examines the impact that speech readability and elegance have on the accuracy of speech-to-text (STT) systems.
Kostek will currently present this knowledge on Thursday, November 21, at 3:25 p.m. What a fascinating topic!
As a pivotal part of the digital 187th Assembly of the Acoustical Society of America, taking place in November… 18-22, 2024.
Medical terminology can be challenging, especially given the varying abbreviations used across different medical specialties? It’s challenging to create an acoustic environment in a hospital setting without proper consultation and planning, as Kostek noted, “That’s a tough job, even more so when we have to navigate real-looking hospital conditions where the room won’t be acoustically ready.”
While Polish remains a primary focus, future plans include expanding linguistic analysis to encompass various languages, including Czech. Collaborative agreements are being forged with the College Hospital in Brno to create novel medical time period resources, leveraging Speech-To-Text (STT) capabilities to enhance healthcare applications.
“While synthetic intelligence excels in various applications, many challenges require a nuanced approach that involves dissecting complex phenomena into their constituent parts.”
