Guides
Why a cry cannot be decoded
Your baby is crying and you want to know why. Here is why that reason cannot be read from the sound of the cry — not as an opinion, but as something we measured.
Agubu confirms a cry from the sound, ranks the likely reasons from context, and says when it is not sure.
On this page
The short answer
A cry carries who the baby is and how old they are. It does not carry why they are crying.
What we measured
We used the only public, commercially usable dataset labelled by cause: 457 recordings from 221 babies. We extracted audio embeddings, asked the same classifier three separate questions, and split the data by baby — one baby's recordings sit either in training or in testing, never both.
That detail decides everything. With the same baby on both sides, the model learns the baby rather than the question, and the score inflates.
Our metric is balanced accuracy. The reason: 83.6% of the recordings carry the label "hungry", so a model that learns nothing and always answers "hungry" already scores 83.6% on raw accuracy.
The result
Chance level in brackets:
- Identity — "which baby is this?" → 62.5% (chance 9.1%), gain +53.4 points
- Age — newborn / infant / older → 40.2% (chance 33.3%), gain +6.9 points
- Cause — three classes → 32.9% (chance 33.3%), gain −0.4 points
- Cause — hungry or not → 49.2% (chance 50.0%), gain −0.8 points
- Cause — pain or not → 51.0% (chance 50.0%), gain +1.0 points
The pipeline works: it tells individual babies apart from their cries. The same pipeline, on the same data, finds nothing about the cause. The method is not the problem. The information is not in the sound.
This is not just our finding
A 2023 study in Communications Psychology examined 39,201 cry sequences recorded in the homes of 24 babies. Across the 676 recordings with a known cause, neither adult listeners nor an algorithm could reliably tell causes apart, while age and identity came through reliably. Our measurement replicates that independently.
Where "96% accuracy" comes from
Of the 221 babies in the dataset, 197 (89%) contributed only a single label — so recognising the baby amounts to knowing the label. Split the recordings randomly instead of by baby and the score rises on its own: on identical data the three-class task went from 32.9% to 35.5%, "hungry or not" from 49.1% to 59.3%, "pain or not" from 50.9% to 62.7%. That is 10 to 12 points on a simple model, and more on a large one.
We also ran a model shipped inside a commercial app on the same data. Given absolute silence, it named a cause with 100% confidence — as it did for all six synthetic inputs we tried. Across different seconds of one recording it changed its answer on more than half the clips, averaging 96.9% confidence while doing so.
Cry detection works; cause detection does not
Two separate questions, and one genuinely has an answer:
- "Is this a cry?" Answerable. Real cries scored a median of 0.737 in our calibration; the hardest synthetic input scored 0.0095, a gap of roughly 78 times. Our threshold is 0.02 — it passes about 90% of real cries and blocks every synthetic input.
- "Why is this baby crying?" Not answerable. That is the list above.
Agubu keeps these apart. The microphone confirms a cry happened; it does not name the reason.
So what does Agubu do
It takes what the sound cannot carry from context: time since the last feed, time awake, the last nappy change, the baby's age, the hour of the day. These genuinely inform.
The result is not an answer but a ranked set of possibilities. When it is unsure it says so, because hiding uncertainty does not remove it — it only hands you misplaced confidence.
When to see a doctor
No app can assess a medical condition behind a cry. See a clinician if:
- Crying lasts more than three hours and nothing settles it
- There is fever, vomiting or a rash
- The cry sounds unusual — high-pitched, weak, or moaning
- Your baby refuses to feed altogether
- The crying started after a fall, a knock or an injury
- Your instinct tells you something is wrong
In an emergency, call your local emergency number.
Sources
Every figure here comes from our own measurements or the named publication. The measurements also appear on the app's "How does it work?" page and are reproducible.
- Our own measurements — donateacry-corpus (457 recordings, 221 babies), baby-wise cross-validation
- Lockhart-Bouron et al., Communications Psychology, 2023 — cries convey age and identity, but not cause
- Our threshold calibration for AudioSet-based cry detection
This article is general information, not medical advice. If your baby has a specific health condition, the instructions your clinician gave you apply.
