Related Experiment Video
Updated: Jan 17, 2026
05:49
Author Spotlight: Deciphering Electrical Networks Behind Complex Brain Activities and Disorders
Published on: November 1, 2024
1.2K
Neonatal pain speech emotion recognition based on Horizontal and Vertical Sparse Mask Time-Frequency Transformer
Jingjie Yan1, Wenjing Sun1, Boyan Sun1
1Nanjing University of Posts and Telecommunications, Nanjing 210003, China.
Computers in Biology and Medicine
|September 18, 2025
Summary
This study introduces a new Neonatal Pain Speech (NPS) database and a HVSMTNet model. The research demonstrates the feasibility of recognizing neonatal pain through speech, outperforming existing methods.
Area of Science:
- Medical research
- Artificial Intelligence
- Speech processing
Background:
- Neonatal pain assessment is crucial but challenging.
- Existing methods lack specialized neonatal pain speech data.
- Accurate pain recognition aids clinical decision-making.
Purpose of the Study:
- To establish a comprehensive Neonatal Pain Speech (NPS) database.
- To develop an advanced model for neonatal pain speech emotion recognition.
- To differentiate pain-induced cries from other vocalizations.
Main Methods:
- Creation of the NPS database with 461 samples across four categories (severe pain, mild pain, crying, calmness).
- Proposal of the Horizontal and Vertical Sparse Mask Time-Frequency Transformer Network (HVSMTNet).
- Implementation of sparse mask mechanisms within time-domain and frequency-domain Transformers.
Main Results:
- The NPS database is effective for distinguishing pain-related cries.
- HVSMTNet achieved superior recognition rates compared to mainstream methods.
- HVSMTNet demonstrated robust performance on the CASIA database.
Conclusions:
- Speech analysis is a viable method for neonatal pain detection.
- The NPS database and HVSMTNet model represent significant advancements in the field.
- This work provides a valuable tool for clinical neonatal care.
