Embryonic Exposure to VPA Influences Chick Vocalisations: A Computational Study

Antonella M. C. Torrisi, Inês Nolasco, Paola Sgadò, Elisabetta Versace, Emmanouil BenetosarXiv 2026 · arXiv · OpenReview

TL;DR

This paper introduces a novel computational framework for analyzing animal vocalizations, moving beyond manual annotation to uncover the complex structure of chick calls. They demonstrate that embryonic exposure to Valproic Acid (VPA), a model for autism, significantly alters chick vocal repertoires, particularly affecting "contact calls," suggesting a call-type-specific vulnerability.

What Problem Does This Paper Address?

Analyzing animal vocalizations, especially in young animals, is crucial for understanding communication, behavior, and developmental health. However, traditional methods rely on manual annotation and predefined categories. This approach is time-consuming, prone to subjective biases, and struggles to capture the full complexity and natural variation within vocal repertoires. This limits scalability and the discovery of subtle differences, particularly in studies of neurodevelopmental disorders where precise characterization is key. The authors aim to overcome these limitations by developing an automated, data-driven framework.

Methods

The core of this work is a computational framework for analyzing chick vocalizations. It consists of three main stages:

  1. Automated Call Detection: Using established algorithms (High Frequency Content for onset, first-order difference for offset) to precisely identify individual vocalizations without manual intervention.
  2. Multi-feature Acoustic Analysis: Extracting 20 one-dimensional acoustic features across temporal, frequency, and energy domains for each detected call. This provides a rich, quantitative description of each vocalization.
  3. Unsupervised Learning (Clustering): Applying Hierarchical Agglomerative Clustering (HAC) to group calls based on their acoustic features. This data-driven approach identifies natural vocal categories without relying on pre-existing human labels, thus avoiding biases.

The framework was then applied to an analytical framework for biological comparisons, using Linear Mixed-Effects Models (LMMs) to statistically assess differences in call production and acoustic features between experimental groups.

Datasets

The study utilized two datasets:

  1. Development Dataset: 31 recordings from newly-hatched, untreated chicks (12-36 hours old). This dataset was used to develop and validate the computational framework.
  2. Experimental Dataset: 46 recordings from chicks exposed in ovo to either Valproic Acid (VPA) or a vehicle control. This dataset, from a previous study [38], was used to test the framework's generalizability and investigate the effects of VPA exposure.

Key Insights and Ideas

  • Vocalizations form a continuum, not discrete boxes: The clustering analysis revealed that chick calls naturally fall into two primary clusters, but these clusters exhibit overlap, suggesting a continuum of vocal behavior rather than strictly separate categories.
  • VPA exposure targets specific call types: The study found that VPA-exposed chicks showed a significant shift in their vocal repertoire, with a relative increase in "softer calls" (Cluster 0) and a decrease in "louder calls" (Cluster 1). This suggests VPA doesn't uniformly disrupt all vocalizations but rather affects specific call types, potentially those related to contact or social interaction.
  • Fine-grained acoustic features reveal subtle changes: By analyzing a comprehensive set of acoustic features, the researchers could pinpoint specific temporal, frequency, and energy modulations in VPA-exposed chicks' calls, highlighting the power of detailed acoustic analysis beyond simple amplitude or frequency measures.

Key Results

  • The computational framework achieved high performance for automated call detection (e.g., weighted F1-measure of 0.933 for onset detection on the experimental dataset).
  • Clustering consistently identified two primary vocal clusters across both datasets.
  • VPA-exposed chicks produced significantly more calls in Cluster 0 (softer calls) and fewer in Cluster 1 (louder calls) compared to controls.
  • VPA exposure led to specific alterations:
  • Temporal: Reduced call duration in Cluster 1.
  • Energy: Quieter Cluster 0 calls and louder Cluster 1 calls in VPA-exposed chicks.
  • Frequency: Reduced pitch variability (e.g., lower F0 Standard Deviation) and altered spectral characteristics in VPA-exposed chicks, particularly in Cluster 1.

Limitations and Future Work

  • Limited recording duration and context: The six-minute recordings on the first day of hatching, during a social preference test, might have elicited specific vocalizations not representative of the entire vocal repertoire or other contexts. Future studies should explore vocal behavior across development and in different contexts.
  • Feature set expansion: While the current feature set is comprehensive, incorporating higher-order features like joint time-frequency scattering coefficients could reveal more complex, nonlinear acoustic interactions.
  • Investigating mechanisms: For Cluster 1 calls, which show greater acoustic variability and F0 modulation, further research is needed to determine if alterations stem from peripheral vocal production, motor impairments, or socio-cognitive deficits.
  • Longitudinal studies: The current study focuses on early development. Longitudinal studies are needed to track whether these vocal alterations persist throughout development.

Summary updated

Abstract

In young animals like poultry chicks (Gallus gallus), vocalisations convey information about affective and behavioural states. Traditional approaches to vocalisation analysis, relying on manual annotation and predefined categories, introduce biases, limit scalability, and fail to capture the full complexity of vocal repertoires. We introduce a computational framework for the automated detection, acoustic feature extraction, and unsupervised learning of chick vocalisations. Applying this framework to a dataset of newly hatched chicks, we identified two primary vocal clusters. We then tested our computational framework on an independent dataset of chicks exposed during embryonic development to vehicle or Valproic Acid (VPA), a compound that disrupts neural development and is linked to autistic-like symptoms. Clustering analysis on the experimental dataset confirmed two primary vocal clusters and revealed systematic differences between groups. VPA-exposed chicks showed an altered repertoire, with a relative increase in softer calls. VPA differentially affected call clusters, modulating temporal, frequency, and energy domain features. Overall, VPA-exposed chicks produced vocalisations with shorter duration, reduced pitch variability, and modified energy profiles, with the strongest alterations observed in louder calls. This study provides a computational framework for analysing animal vocalisations, advancing knowledge of early-life communication in typical and atypical vocal development.
BibTeX
@misc{torrisi2026embryonic,
  author = {Antonella M. C. Torrisi and Inês Nolasco and Paola Sgadò and Elisabetta Versace and Emmanouil Benetos},
  title = {{Embryonic Exposure to VPA Influences Chick Vocalisations: A Computational Study}},
  year = {2026},
  eprint = {2601.12203},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2601.12203},
}
Figure