Parrots / Neural and bioacoustic study with a separate data deposit

Budgerigar vocal signatures: caller identity, not call meaning

Source year 2023 · Note reviewed 2026-10-14

The full article used 16 adult budgerigars, with eight in the contact-call cohort and five in the neural-inactivation experiment. Playback of colony-noise segments elicited 180–1,612 contact calls per bird per day in the eight-bird cohort. A VAE representation and SVM identified callers at over 90% accuracy in a balanced random 80/20 call-level split, repeated 50 times; accuracy remained about 90% for two shared call types. A classifier trained at week 3 and tested at week 15 fell from about 90% to 40%, consistent with call change over time. In the separate five-bird experiment, birds still responded to individual calls during AFP inactivation, although response rate fell in three birds and latency effects varied. Inactivation altered call acoustics and reduced individual discriminability. These are bounded acoustic identity and production findings, not decoded meaning.

What this does — and doesn’t — establish

The reported identity classifier randomly split calls from the same known birds; no held-out-bird test is reported, so its accuracy is not generalization to unseen animals, species, sites or devices. The study used small captive groups, controlled loudspeaker playback, directional laboratory microphones and invasive neural interventions. Receiver responses were measured in five birds and were not a test of name comprehension or call meaning. The official data page associates ColonyNoiseCalls with Figure 2, the eight-bird contact-call identity analysis; the single inspected file has no verified mapping to a specific bird, session or condition and no independent behavior-label field. The paper's reviewed methods do not state a digitization rate, so the sample's stored 44,101 value remains unexplained. The dataset separately declares CC BY 4.0; custom scripts are available from the lead contact without a verified public license.

What we can build on

Treat caller identity, receiver response, context association and communicative meaning as different outcomes. A first identity benchmark would need bird-grouped holdouts; a meaning claim would need independently measured receiver behavior. The specialized playback and neural methods do not validate a phone workflow or justify app playback.

See the related tools and reading paths →

The next reading task

If the deposit is considered for a defined analysis, obtain the versioned per-file manifest and map this sample to caller, session and condition; resolve the sample rate against acquisition metadata rather than guessing from the filename. The official file-list API requires OAuth2 and the unauthenticated request returned 401. Verify exact file terms and obtain independent avian-bioacoustics/statistical review before any benchmark. Until then, keep this study as a methods comparator, not a ready context or translation benchmark.

Go to the sources

1. Budgerigar vocal signatures and neural pathways ↗

   Zhao et al. (2023), Current Biology 33(24):5415–5426.e4. Full article text reviewed through NCBI PMC E-utilities on October 14; the identity classifier used random call-level splits within known birds.

2. Budgerigar study data deposit ↗

   Zhao (2023), version 1, CC BY 4.0. The landing page says the MATLAB struct-array deposit includes raw audio, with Figure 2 data in ColonyNoiseCalls and Figures 3–5 data in acoustiSAT_MAO.mat.

3. Mendeley Data API documentation ↗

   Official documentation for public dataset file-list and file-detail endpoints, with OAuth2 authentication; OAI-PMH is documented for public-record metadata harvesting.

Original editorial note by Talk2Nature contributors. AI-assisted, source-grounded synthesis; not a peer review or an independent replication. Linked works retain their own rights.