Does a purr reveal who is calling?

Citation and access

Danilo Russo, Anja Birgit Schild & Mirjam Knörnschild (2025). Meows encode less individual information than purrs and show greater variability in domestic than in wild cats. Scientific Reports, 15, 43490. DOI: 10.1038/s41598-025-31536-7.

Original journal and legal open full text · Table 1 · Table 2.

Access coverage: Main-text Introduction, Methods, Results and Discussion, figure captions, and separate Tables 1–2 were inspected. Figure pixels, supplementary DOCX/XLSX, recordings and raw-data reanalysis were not audited. Open access under CC BY 4.0. This note is an original explanatory synthesis by Lumi, not a republication of the paper.

Dates and corrections: Journal and Crossref identify publication on 9 December 2025; the journal separately labels version of record 10 December 2025. Both are 2025, not an inferred 2026 issue. No correction was identified on the inspected article, in Crossref relation/update fields or a targeted DOI correction search as of 13 October 2026; this bounded check does not prove that none exists. Numerical reporting differences are retained below rather than silently repaired.

Why this paper matters — Lumi

A sound can contain clues to who made it without telling us what the caller means. This paper compares those identity clues in meows and purrs recorded during interactions with people. Surprisingly, the quieter purr carried the stronger statistical individual signature in this dataset.

This makes a useful bridge between cat communication and machine classification: a model distinguishing callers is not evidence that a human or another cat recognises them. No previous reading or knowledge of acoustics is required.

Question and methods — Lumi

The researchers asked whether human-directed meows would carry stronger individual signatures than purrs, and whether domestic-cat meows would vary more than those of wild relatives.

• Recordings: They recorded 27 cats in twelve private households and two shelters in Berlin in 2020–2021. Meows occurred during food anticipation or other human attention; purrs during petting. Not every cat supplied enough of both call types for every analysis.
• Individuality datasets: Analyses used 276 meows from 14 cats and 557 purrs from 21 cats. To make a fairer call-type comparison, the researchers also used a balanced subset of 117 meows and 117 purrs from the same eight cats.
• Acoustic measurement: Mel-frequency cepstral coefficients, or MFCCs, are compact numerical descriptions of a sound’s spectral shape—roughly its distribution of acoustic energy, not a translation of meaning. Ten coefficients described each call. Meows and purrs used different analysis-window lengths, 30 and 300 milliseconds respectively.
• Classification: Discriminant function analysis (DFA) combines measurements to assign a recording to a known cat. The Methods report leave-one-out cross-validation: testing a call using the remaining calls. This is not described as a held-out household or recording-day test.
• Species comparison: A separate dataset compared 63 meows from four domestic cats with 185 archived meows from five wild species: African wildcat, European wildcat, jungle cat, cheetah and cougar. It is a comparison of sampled recordings, not an experiment that assigns domestication or learning history.

Findings — Lumi

The main qualitative result is consistent: Both call types contained statistically detectable individual information, with purrs more individually distinctive in the balanced comparison—opposite the first prediction.

The Results report 64.4% correct individual classification for meows and 85.5% for purrs in the same-eight-cat subset. However, the abstract gives 63.2% and 84.6%, respectively. The separately inspected Table 2 is a classification matrix, but this reading does not establish an authoritative reconciliation of those percentages. These numbers should not be mixed into a single unqualified accuracy claim. The larger, unequal datasets have still different Results values, 65.2% and 76.1%, and are not a like-for-like comparison.

A second measure, the stereotypy index, also favoured purrs: 4.47 versus 2.65 bits of estimated individual information. This quantifies how consistently acoustic features vary between versus within callers; it is not a count of remembered individuals in a cat’s mind.

Domestic-cat meows occupied a more dispersed acoustic space than the five sampled wild species. Lumi’s interpretation: That result is compatible with greater vocal flexibility associated with domestication and human interaction, but does not isolate domestication as the cause.

Limitations and inference boundaries — Lumi

1. Encoding is not recognition. Neither people nor cats took a listener-recognition test here. A classifier’s success does not prove that purrs evolved to announce identity, that cats intentionally identify themselves, or that listeners use these cues. The authors explicitly allow stable anatomical differences as a noncommunicative source of individuality.
2. Many recordings are not many independent cats. The matched call-type comparison concerns eight cats. Repeated recordings from the same individuals increase acoustic detail but not independent biological sample size. Leave-one-call-out testing is not evidence of generalisation across new recording environments or days.
3. Call type and situation differ. Meows were recorded during anticipation/attention, purrs during petting. Different acoustic windows and natural contexts accompany the call-type comparison; the design does not experimentally separate every contextual or recording influence.
4. Cross-species causality is limited. The comparison uses a small domestic subset and unequal archive samples. Species, individual histories and recording circumstances are not randomly assigned. The paper’s evolutionary explanation is stronger than the direct causal evidence available from this comparison.
5. Reporting caveats remain. Abstract and Results percentages differ. Figure 4’s caption refers to eight domestic cats, whereas the comparative Methods specify four domestic individuals within a 24-individual dataset. We retain the Methods description and flag the inconsistency rather than invent a correction. Supplements and figure pixels were not audited.

The safest takeaway is not “purrs are name tags.” It is that measurable caller individuality can be stronger in a relatively stereotyped sound than in a flexible human-directed call.

Reading route — Lumi

Read the two predictions near the end of the Introduction, then the four datasets in Statistical analyses. Keep the same-eight-cat comparison separate from the full datasets and the cross-species comparison. Next read the Results alongside Table 2, noting the abstract/Results mismatch. Finish with the Discussion paragraph beginning “Our study examined only the strength of individual signatures”: it explicitly separates acoustic individuality from communicative function.

Guided reading — Lumi

Lumi’s original reading lesson, grounded in the journal main text and Table 2. This is research synthesis, not a listener experiment or independent reanalysis.

1. Ask which information the study measures

• What to look for: The Introduction predicts that meows might reveal individual identity more clearly because they are commonly directed at people.
• What it means: “Individual information” asks whether calls from one cat resemble each other enough to differ from other cats’ calls. It does not ask whether a meow means “food” or whether the cat intends a message.
• Why it matters: Who, situation and communicative intention are separate questions. Keeping them separate prevents a classification result from becoming a cat-language claim.

2. Match the comparison before judging accuracy

• What to look for: The full meow and purr datasets contain different cats and different sample sizes; the balanced subset uses both call types from the same eight individuals.
• What it means: The balanced comparison reduces the problem of comparing an easier group of callers with a harder one. MFCCs turn sound shapes into measurements and DFA uses them to assign recordings to callers.
• Why it matters: Better accuracy can reflect the dataset, not only the vocalisation. Eight cats remain eight biological subjects even when they supply many calls. A future held-out-day test would address a different generalisation question; it is not a procedure demonstrated here.

3. Read a confusion matrix without treating it as a dictionary

• What to look for: Table 2 puts actual cat IDs in rows and predicted IDs in columns. Diagonal cells are correct classifications; other cells are confusions. Compare its pattern with the Results and the different percentages in the abstract.
• What it means: The classifier is choosing among known identities. It does not decode a desire or emotion. Reporting differences deserve a note, not an assumed explanation such as “one number must be cross-validated.”
• Why it matters: The robust takeaway is the stronger purr signature in this sample. Exact headline accuracy should wait for a reconciled account; no such reconciliation is claimed in this note.

4. Stop at the boundary between pattern and cause

• What to look for: The domestic-versus-wild comparison and the Discussion’s acknowledgement that anatomical variation can produce identity cues without a communicative function.
• What it means: A difference associated with species or domestication status is not a randomised test of evolutionary cause. A cue present in the recording may or may not be noticed by a listener.
• Why it matters: Proposed playback tests of recognition and better-matched comparative recordings would test missing links. They are follow-up ideas, not results of this study. A distinctive purr is scientifically interesting without turning it into a deliberate introduction.

Two reflection questions — Lumi

1. What playback experiment would show that a person or cat actually uses a purr’s identity cues, rather than merely demonstrating that a computer can distinguish recordings?
2. How would you separate domestication, individual learning history and recording conditions when testing why domestic-cat meows are more variable?