Related Experiment Video
Updated: May 5, 2026
10:28
Recording Mouse Ultrasonic Vocalizations to Evaluate Social Communication
Published on: June 5, 2016
25.7K
Hierarchical Acoustic Encoding Distress in Pigs: Disentangling Individual, Developmental, and Emotional Effects with
Irenilza de Alencar Nääs1, Danilo Florentino Pereira2, Alexandra Ferreira da Silva Cordeiro1
1Graduate Program in Production Engineering, Paulista University, Rua Dr. Bacelar 1212, São Paulo 04026-002, SP, Brazil.
Animals : an Open Access Journal From MDPI
|May 4, 2026
Summary
This study developed a pig vocalization analysis method to monitor animal welfare, achieving 60.9% accuracy in identifying distress states across different pigs and growth phases. Subject-wise validation ensures reliable, generalized welfare monitoring in Precision Livestock Farming.
Area of Science:
- Animal Science
- Bioacoustics
- Precision Livestock Farming
Background:
- Scalable, non-invasive methods are crucial for automated pig welfare monitoring across diverse individuals and life stages.
- Individual vocal differences and growth-related changes can complicate the accurate interpretation of pig distress vocalizations.
- Subject-wise validation is essential to ensure monitoring systems generalize to unseen animals, preventing inflated accuracy.
Purpose of the Study:
- To develop and validate a subject-wise approach for analyzing pig vocalizations to monitor welfare states.
- To quantify the influence of distress, growth phase, sex, and individual identity on acoustic features.
- To establish a benchmark for acoustic distress detection in pigs and explore limitations for future multimodal systems.
Main Methods:
- Analyzed 2221 vocal samples from 40 pigs across four growth phases and six conditions (pain, hunger, thirst, stress, normal).
- Extracted acoustic features (energy, duration, intensity, pitch, formants) using Praat software.
- Employed mixed models with animal identity as a random effect and grouped cross-validation (5-fold) for subject-wise testing.
Main Results:
- Distress exposure significantly impacted intensity and spectral traits (Formant 2); growth phase affected duration and pitch.
- Animal identity contributed a modest but consistent variance (R² ~0.02-0.03) beyond other factors.
- A Random Forest model achieved 60.9% balanced accuracy and 59.7% macro-F1, with pain being the most distinguishable state.
Conclusions:
- Hierarchical acoustic encoding of distress is supported, establishing a benchmark for precision welfare monitoring.
- Subject-wise validation is critical for accurate generalization in automated welfare assessment.
- Future systems require multimodal integration (bioacoustics, environmental, behavioral data) to resolve complex physiological overlaps.
