Samsung R&D Institute Poland Engineers Shine at INTERSPEECH 2025 Challenges and Conference

The AI Solutions team at Samsung R&D Institute Poland (SRPOL) achieved outstanding results at INTERSPEECH 2025, securing 1st place in the Mice Autism Detection via Ultrasound Vocalisation (MADUV) Challenge, 2nd place in the Speech Emotion Recognition in Naturalistic Conditions Challenge, and having five papers accepted for presentation at the conference.

Challenge Results:

    
·
First place in the MADUV Challenge, a pioneering task on classifying mice as wild-type or autism model based on ultrasound vocalisations
    
·
Second place in the Speech Emotion Recognition in Naturalistic Conditions Challenge, covering categorical and attribute-level emotion prediction from spontaneous podcast speech


[Winning Team of AI Solutions team, SRPOL]

INTERSPEECH 2025 Papers from SRPOL:

SRPOL researchers also had five papers accepted at INTERSPEECH 2025:

    
1.
Clustering-based Hard Negative Sampling for Supervised Contrastive Speaker Representation Learning – P. Masztalski, M. Romaniuk, J. Żak, M. Matuszewski
    
2.
Significance of Time-Frequency Preprocessing for Automatic Ultrasonic Vocalization Classification in Autism Spectrum Disorder Model Detection – S. Szmajdziński, J. Kruk, I. Ryzhankow, L. Łazarski, J. Żak, W. Średniawa
    
3.
Multi-task Learning for Speech Emotion Recognition in Naturalistic Conditions – B. Zgórzyński, J. Kruk, P. Masztalski, W. Średniawa
    
4.
Leveraging Text and Speech Processing for Suicide Risk Classification in Chinese Adolescents – J. Krzywdziak, B. Eljasiak, J. Stępień, M. Świątek, A. Pruszek
    
5.
TinyClick: Single-Turn Agent for Empowering GUI Automation – P. Pawlowski, K. Zawistowski, W. Lapacz, A. Wiacek, M. Skorupa, S. Postansque, J. Hoscilowicz


Driving Health & Well-Being Through AI

These accomplishments underscore SRPOL’s mission to advance Samsung’s R&D capabilities in health and wellbeing through cutting-edge speech and machine learning research. By developing AI solutions for early detection of developmental disorders, accurate recognition of human emotions, and identification of mental health risks, SRPOL is contributing to innovative tools that can support healthcare professionals, improve patient outcomes, and enhance quality of life worldwide.