Faculty of Information Technology, BUT（https://www.fit.vut.cz/.en）
News
Category: news
Day: 29 October 2026
When good test results aren’t enough: Hanan Aldarmaki discusses the limitations of speech recognition
Automatic speech recognition is currently achieving very good results in tests. But does that mean it works just as well in real-world applications? On Tuesday 3 November at 1.00 pm , we will welcome Hanan Aldarmaki from the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) to FIT, Brno University of Technology. In her lecture, Aldarmaki will highlight the fact that even seemingly minor errors in speech recognition systems can pose a significant problem in practice.
“Beyond Benchmarks: What Real Applications Reveal About Speech Recognition”
📅 Tuesday 3 November 2026, 13:00
📍 FIT VUT, E105
The lecture will focus on the difference between the results of automatic speech recognition in standard benchmark tests and its performance in specific applications . Using examples from clinical speech and language assessments and educational games, Hanan Aldarmaki will present situations in which systems encounter, for example, a lack of linguistic context or the difference between what a person actually said and what they probably intended to say.
It is precisely context that can significantly improve recognition; at the same time, however, it can lead the system to ‘correct’ incorrect pronunciation or speech disfluency. Yet these are precisely the phenomena that may need to be captured, for example, during a clinical examination. The lecture will also present experiments investigating the influence of context, accent and the recognition of isolated words, and will raise the question of how to develop and evaluate speech recognition systems according to the needs of their actual users.