Vocal repertoire of adult domestic pigs in a laboratory environment

This study characterizes the vocal repertoire of adult domestic pigs in a laboratory setting by analyzing 1,136 vocalizations to define five major call classes and 16 distinct types, thereby establishing a detailed acoustic framework to support non-invasive welfare assessment and comparative research.

The Gist

Imagine you walk into a busy office building. You hear a mix of sounds: the low hum of conversation, the sharp ring of a phone, the rhythmic tapping of keyboards, and the occasional shout of frustration. If you've never been in that office before, you might just hear "noise." But if you work there, you know exactly what each sound means: the phone ringing means a client is calling, the shouting means a deadline is missed, and the low hum is just the background chatter.

Now, imagine that office is filled not with humans, but with pigs.

For decades, scientists have used pigs to study human diseases because they are smart, social, and their bodies work very similarly to ours. But until now, we've been like that new visitor in the office: we could hear the pigs, but we didn't really understand their "language." We knew they made noise, but we didn't have a dictionary to translate it.

This paper is essentially the first comprehensive dictionary and grammar guide for the voices of adult pigs living in a research lab.

Here is the story of how they did it, explained simply:

1. The Mission: Decoding the "Oink"

The researchers wanted to know: What are these pigs actually saying?
In the wild, or on a farm, pigs make sounds to talk to each other. But in a sterile, concrete-walled laboratory, surrounded by humans in white coats, do their voices change? Do they have a different "office slang"?

To find out, they recorded 12 pigs (a mix of boys and girls) over several weeks. They recorded them during normal daily life, when scientists walked in to feed them, when they were getting a gentle touch, and even when they were being led on a leash. They captured over 1,100 distinct sounds.

2. The Two Detective Teams

To make sense of this mountain of audio, the researchers used two different detective methods, like having a human expert and a super-computer work together.

• Team Human (The "Ear and Eye" Method):
  A researcher listened to the sounds and looked at the visual waveforms (like a fingerprint of the sound). They acted like a music critic, grouping sounds by how they felt.

  • The Result: They found 5 main "genres" of pig voices:
    1. Grunts: The low, rumbling "hello" or "I'm hungry" sounds.
    2. Squeals: The high-pitched, noisy sounds.
    3. Screams: Long, loud, high-pitched cries (often of distress).
    4. Complex: A mix of two sounds at once (like a grunt that suddenly turns into a squeal).
    5. Barks: Short, sharp bursts.
  • They then broke these down further into 16 specific "dialects" (like a "classic grunt" vs. a "growl").

• Team Robot (The "Math" Method):
  They fed the data into a computer program that looked at the numbers: How long did it last? How high was the pitch? How wide was the frequency? The computer tried to sort the sounds into groups without any human bias.

  • The Result: The computer said, "Okay, I see two main groups: Low Frequency and High Frequency."

3. The Big Discovery: It's More Complicated Than Two Groups

Here is the twist. The computer was right that there are two main groups, but it was too simple. It was like a computer trying to sort a library into just "Fiction" and "Non-Fiction." Sure, those are the two main categories, but inside "Fiction," you have mysteries, romances, sci-fi, and horror. They are all fiction, but they are very different!

The researchers found that while the math said "two groups," the human ear could hear much more nuance. The pigs have a rich, complex vocabulary.

• They discovered new words the pigs use that nobody knew about before, like "Whines" (a sad, harmonic sound) and "Yelps" (a sharp, high-pitched noise).
• They found that the pigs in the lab had developed specific ways of talking that might be unique to their concrete surroundings, different from pigs on a farm or in the wild.

4. Why This Matters: The "Honest Signal"

Why do we care if a pig grunts or squeals?
Because pigs are honest signalers. They can't lie.

• If a pig is stressed, in pain, or happy, their voice changes in a specific way.
• Before this study, if a scientist heard a pig squeal, they might just think, "Oh, the pig is making noise."
• Now, thanks to this "dictionary," a scientist can say, "That's a Stable Scream," which tells us the pig is likely in significant pain or distress. Or, "That's a Modulated Grunt," which might mean the pig is just curious or greeting a friend.

The Takeaway

Think of this paper as the Rosetta Stone for pig communication in the lab.

By creating a standard list of what these sounds mean and how they sound, the researchers have given scientists a new tool. Instead of just guessing if a pig is happy or sick, they can now "listen" to the pig's voice to check its health. This helps ensure that the pigs used to help cure human diseases are treated with the utmost care and respect.

In short: We finally learned how to listen to the pigs, and it turns out they have a much more interesting conversation than we ever imagined.

Technical Summary

1. Problem Statement

The use of domestic pigs (Sus scrofa domesticus) as a translational model for human biomedical research has increased eightfold in the last three decades. Despite their value due to genetic heterogeneity and physiological similarity to humans, there is a critical lack of reliable, non-invasive methods to assess their health and welfare in laboratory settings. While vocalizations are known "honest signals" of an animal's internal state (reflecting stress, pain, and emotion), the acoustic repertoire of adult domestic pigs in biomedical laboratory environments remains poorly defined.

Existing literature primarily focuses on:

• Piglets or wild boars.
• Agricultural settings (open barns) rather than indoor concrete laboratory facilities.
• Perceptual analysis without robust objective statistical validation.

The authors hypothesize that the unique acoustic properties of laboratory housing (reverberation, echo) and the specific human-animal interactions in research settings may result in a vocal repertoire distinct from those previously characterized in agricultural or natural environments.

2. Methodology

Subjects and Environment:

• Subjects: 12 mixed-breed domestic pigs (5 males, 7 females), aged 2–3 months (11–13 kg).
• Housing: Single pens in an indoor facility with concrete walls, controlled temperature (72±2°F), and 12-hour light/dark cycles.
• Protocol: Recordings were taken during two phases:
  1. Pre-acclimation: Upon arrival (Day 1–2).
  2. Post-acclimation: After a 2–3 week habituation period involving positive reinforcement and leash training.

Data Collection:

• Contexts: Recordings occurred during 11 distinct scenarios, categorized into Laboratory Procedures (investigator approach, touch, board constraint, intramuscular injection, food presentation, leashed walks) and Social Interactions (de-crating, separation/reintroduction, hierarchy establishment).
• Equipment: Beyerdynamic MCE 86 S II shotgun microphone and GoPro Hero 2 camera. Audio sampled at 48 kHz, 16-bit.
• Dataset: 1,174 vocalizations were initially identified; 1,136 high-quality vocalizations (from 69 recordings) were retained after excluding overlaps and excessive echo.

Analysis Techniques:
The study employed a dual-approach classification strategy:

1. Aural-Visual (AV) Classification:
   • An investigator, blind to the context and sex of the animal, analyzed spectrograms generated via Raven Pro 1.5.
   • A hierarchical structure was built: Vocal Classes $\rightarrow$ Subclasses $\rightarrow$ Call Types.
   • Robust acoustic measures were calculated to minimize manual selection bias: 5% frequency ($f_5$), 95% frequency ($f_{95}$), First/Third Quartile frequencies ($Q_1f, Q_3f$), Center Frequency ($Cf$), 90% Bandwidth ($BW_{90}$), Interquartile Range Bandwidth ($IQRBW$), and 90% Duration ($DUR_{90}$).
2. Objective Cluster Analysis:
   • A two-step cluster analysis was performed using IBM SPSS Statistics.
   • Variables were converted to z-scores. Multicollinearity was checked (Pearson $r > 0.90$).
   • The optimal number of clusters was determined using Schwarz's Bayesian Information Criterion (BIC) and distance measure ratios.
3. Statistical Validation:
   • Agreement: Cramer's V was used to measure agreement between the AV classification and the cluster analysis.
   • Independence: A mixed linear model assessed whether individual pig identity significantly influenced cluster membership (random effect).

3. Key Contributions

• First Comprehensive Repertoire for Adult Lab Pigs: This study provides the first detailed acoustic framework specifically for adult domestic pigs housed in biomedical research facilities, distinguishing them from agricultural or wild populations.
• Standardized Definitions: The authors established standardized, quantitative definitions for 16 distinct call types, integrating descriptive labels with specific acoustic thresholds. This creates a "common language" for researchers.
• Novel Call Types: The study identified call types not previously reported in adult domestic pigs, including whines, yelps, and stable screams.
• Methodological Rigor: By combining perceptual (AV) analysis with unsupervised machine learning (two-step clustering), the study validates subjective classifications with objective statistical data.

4. Key Results

Hierarchical Structure (AV Classification):

• 5 Vocal Classes: Grunts, Squeals, Screams, Complex Vocalizations, and Barks.
• 16 Call Types:
  • Grunts: Classic, Croak, Growl, Modulated, Short.
  • Squeals: Classic, Yelp, Whine, Squeak.
  • Complex: Complex Squeal, Complex Yelp, Complex Whine, Complex Scream.
  • Screams: Modulated, Stable.
  • Barks: (Single type).
• Distribution: Grunts were the most dominant (81.95% of all vocalizations), followed by Squeals (10.5%), Complex (4.5%), Screams (2.3%), and Barks (0.6%).

Acoustic Characteristics:

• Grunts/Barks: Lowest frequency, shortest duration, pulsatile structure.
• Screams: Highest frequency, longest duration, high frequency modulation.
• Complex Calls: Typically began as a grunt and transitioned into a squeal or scream.
• Unique Findings: "Whines" were distinguished by short-duration harmonics and lower frequencies than other squeals. "Stable screams" were defined by long duration and low frequency modulation, contrasting with "modulated screams."

Cluster Analysis:

• Optimal Solution: The two-step algorithm identified 2 clusters as the optimal statistical solution (BIC = 4325.82).
  • Cluster 1: 95.9% of cases; characterized by lower frequencies.
  • Cluster 2: 4.1% of cases; characterized by higher frequencies.
• Predictors: Center Frequency ($Cf$) and First Quartile Frequency ($Q_1f$) were the most important predictors for cluster membership.
• Nuance: While the 2-cluster solution was optimal, the BIC values showed secondary peaks at 5 and 19 clusters, suggesting that while the data is heavily skewed toward low-frequency sounds, a richer acoustic structure exists that standard clustering struggles to resolve due to gradation between call types.

Agreement and Independence:

• Agreement: There was moderate-to-high agreement between the perceptual AV classification and the objective cluster analysis (Cramer's V = 0.67, $p < 0.0001$).
• Individual Variation: Individual pig identity explained only a small, non-significant percentage of the variance in cluster membership (approx. 3-3.6%), indicating that the vocal repertoire is consistent across individuals in this environment.

5. Significance

• Welfare Assessment: The detailed acoustic framework allows for the development of predictive models to assess pig welfare non-invasively. Specific call types can now be linked to specific stressors (e.g., pain during injection vs. social stress).
• Translational Research: As pigs become a primary model for human disease, understanding their behavioral endpoints is crucial. This study ensures that behavioral data (vocalizations) collected in labs are interpreted accurately, reducing confounding variables in experimental outcomes.
• Future Directions: The authors suggest that while binary (high/low frequency) clustering is statistically dominant, future work should employ "fuzzy" clustering methods to better capture the gradation between call types. Additionally, the standardized definitions provided here enable cross-laboratory comparisons, essential for meta-analyses in translational science.

In summary, this paper bridges the gap between ethological observation and objective acoustic analysis, providing a robust, standardized vocabulary for the vocal behavior of adult laboratory pigs, which is essential for advancing animal welfare and the reliability of biomedical research.