The most striking welfare result published on broiler houses in recent years does not come from a weight, a temperature or a gait score. It comes from how much the chicks call during their first four days on the floor, and from what that noise turns out to predict four weeks later.
Key takeaways
- Spectral entropy per minute of recording correlated at −0.88 with a manual distress call count, and the relationship held independently of age over days 1 to 4 (Herborn et al., 2020).
- The fundamental frequency of broiler calls falls from around 3 kHz in the first week to 1.2 to 2.0 kHz by week 5, so any alarm threshold that ignores bird age will drift.
- A published sneeze detection algorithm on birds aged 15 to 45 days reached a sensitivity of 66.7 % and a precision of 88.4 %. Specificity is high, sensitivity is the weak point.
- No part of EU broiler welfare law uses sound. Directive 2007/43/EC still rests on stocking density and on mortality data, so acoustics is a management tool, not a compliance one.
What a good stockperson already hears
Anyone who has spent a season in a broiler house can tell a settled flock from an unsettled one before switching on a light. The change is not in volume alone but in texture: a contented flock produces a broad, low, overlapping murmur, while a stressed one produces repetitive, high energy calls that cut through it. That perception is real, and it is the starting point of every automated system now being tested.
The problem with the trained ear is not accuracy, it is coverage. It is present for a few minutes per house per day, it does not record, and it cannot compare Tuesday with the Tuesday three crops ago. Everything that follows in this article is an attempt to keep the same signal and remove those three limitations.
The day 4 result, and why it matters commercially
Herborn and colleagues recorded 12 commercial Ross 308 flocks over the four days following placement and used spectral entropy, a measure of how evenly sound energy is spread across frequencies, as a proxy for distress calling. High pass filtered spectral entropy tracked manual distress call counts closely, with a Spearman correlation of −0.88, and the slope was consistent across days 1 to 4.
The predictive part is the commercially interesting one. Low median daily spectral entropy on day 4, meaning a lot of distress calling, predicted higher cumulative flock mortality at day 32 (slope −0.171 ± 0.067, p = 0.028) and lower average bird weight at day 32 (slope 2.32 ± 1.00, p = 0.043). Low entropy also predicted higher mortality the following day, on every measurement day.
Read operationally, that turns brooding noise into an early indicator rather than a description. It does not tell you what is wrong. It tells you, on day 4, that this crop is on a worse trajectory than the last one, at a point where temperature, litter, drinker height and chick distribution can still be corrected.
Why an alarm threshold has to be age corrected
Bird calls change as the bird grows, and the change is large enough to break a fixed threshold. Published work on broiler vocalisation reports the fundamental frequency of calls dropping from roughly 3 kHz in the first week to between 1.2 and 2.0 kHz by week 5, depending on the call type, with distress calls and short peeps showing a clearly descending pattern that flattens as the birds age. Vocalisation also varies with time of day, which means a system compared against yesterday at the same hour behaves very differently from one compared against a daily average.
Thermal environment adds a second correction. Chicks kept in thermal comfort vocalise less than chicks under thermal stress, and vocal activity rises when temperature moves away from the optimal brooding range. A rise in calling is therefore a signal to check the environment first, before looking for a health explanation. The same reading discipline applies to visual and postural cues, which is why we treat acoustics as one input among several in reading poultry behaviour and its signals rather than as a standalone system.
Where automated detection still breaks down
Distress calling is loud, frequent and spread across the flock, which is exactly why it was the first signal to be measured reliably. Respiratory sounds are the opposite. An algorithm developed to detect sneezes in broilers aged 15 to 45 days, trained on 763 annotated sneezes from 480 minutes of recording in a group of 51 birds, reached a sensitivity of 66.7 % with a precision of 88.4 %.
Reviews of the field are blunter still. Much of the published work was carried out under laboratory conditions with background noise deliberately excluded; specificity for sneezing reached almost 100 % while sensitivity stayed low, which makes the technique ineffective at catching true positives in a working house. Part of the problem is arithmetic rather than engineering: subtle sounds are rare, in one study only 0.24 % of recorded sound, so there is very little labelled data to train on.
| Acoustic signal | What published work associates it with | Current practical limit |
|---|---|---|
| Early life distress calling | Cumulative mortality and average weight at day 32 | Predicts a trajectory, identifies no cause |
| Call frequency shift with age | Normal development, from about 3 kHz to 1.2 to 2.0 kHz | Requires age corrected baselines |
| Vocal activity under heat or cold | Departure from the thermal comfort range | Confounded with density and light events |
| Sneezing and rales | Respiratory disease | Sensitivity around 67 % in trial conditions, lower in noise |
Fitting a microphone into a house that already has sensors
Three practical points decide whether an acoustic trial produces anything usable.
- Placement and noise floor. Fans, feed lines and augers dominate the spectrum. A microphone that shares a wall with a fan bank measures the fan, and a system commissioned in winter ventilation will behave differently in summer.
- Baselines per house, not per company. Building height, litter, equipment and flock size all change the acoustic picture, so a threshold imported from another site starts wrong.
- Recording in a workplace. A microphone in a poultry house also captures staff conversation. Before installing one, check with whoever handles data protection on your holding what is retained, for how long and on what basis, because that question does not disappear simply because the target of the recording is a bird.
None of this replaces the monitoring that the law already requires. Under Directive 2007/43/EC, maximum stocking density is 33 kg per square metre, rising to 39 kg per square metre where the Annex II conditions on ventilation, temperature, humidity, ammonia and carbon dioxide are documented, and up to 42 kg per square metre in the exceptional circumstances defined in Annex V, subject to mortality performance. Daily and cumulative mortality accompany the flock to the slaughterhouse. Acoustic data sits alongside that record; it does not substitute for it.
Three questions operators put to us
Can sound analysis replace a house visit?
No, and treating it that way is the fastest route to disappointment. What it does is tell you which house to walk into first, and which day of the crop is deviating from the last one. The diagnosis still happens on the floor.
Is this technology ready for commercial purchase?
It depends on the signal. Distress call analysis has been validated on commercial flocks at scale. Automated respiratory sound detection has not reached comparable reliability under real house noise, so a supplier claiming disease detection should be asked which signal, on which flock size, at what sensitivity, and in what background noise.
What should we do if calling rises sharply?
Check the environment before anything else: temperature at chick level, drinker access, litter condition, draughts and light schedule. If the environment is in order and the calling persists, that is a conversation for the veterinarian responsible for the flock, not for the sound system’s dashboard.
Sound is one instrument among several
Light, sound and air quality are being reworked together in modern facilities, and the gains usually come from how they are combined rather than from any single sensor.
Sources: Herborn et al., Spectral entropy of early life distress calls as an iceberg indicator of chicken welfare, Journal of the Royal Society Interface, 2020; published research on the influence of age, time of day and environmental changes on vocalisation patterns in broiler chickens; Development of a sound based poultry health monitoring tool for automated sneeze detection, Computers and Electronics in Agriculture, 2019; systematic review on the detection of respiratory diseases through precision farming; Council Directive 2007/43/EC laying down minimum rules for the protection of chickens kept for meat production, Annexes II and V. Consulted August 2026.
Published previously, fully revised on 17 August 2026. General operational guidance for poultry professionals. It does not replace the assessment of the veterinarian responsible for your flock, nor the requirements applicable to your holding under national implementation of EU welfare rules.