Birds / Bioacoustic identity study

Monk parakeet voiceprints test caller identity, not meaning

Source year 2023 · Note reviewed 2026-10-13

A study of 229 monk parakeets tests caller identity across call types; it does not predict behavior or decode meaning.

What we reviewed · Targeted study and dataset-fit review

Reviewed the final-article abstract and targeted methods/figure text surfaced by the primary PMC index, plus the official pinned GitHub README and the existing hash-pinned 2020 annotation sample. Recomputed the sample's Git blob SHA-1 and SHA-256; the Git blob matches the pinned GitHub file. Rechecked Edmond's released v1 API, current Data Access & Usage guide, ZIP-viewer FAQ, Terms of Use and official July 20, 2026 terms-update notice. No Edmond terms were accepted and no deposit archive/member was downloaded. Direct full-text retrieval was CAPTCHA-limited; the ANU repository PDF fetch timed out. The supplement, full statistical/code workflow and raw audio were not reviewed or run.

Not yet independently reviewed by a scientific specialist.

Study at a glance

Research question: Are monk-parakeet callers individually distinguishable within and across call types? The target is caller identity, not call meaning or behavior.
Study sample: 5,599 vocalizations from 229 individually marked wild monk parakeets in Barcelona, recorded across 2020–2021. The authors categorized 11 call types and analyzed five with enough observations.
Reported result: The final journal abstract reports individual signatures in three of five call types, rapid variation in contact calls, and evidence that a discriminant function predicts caller identity across call types.
Cross-call test: Figure 4 describes splitting recordings into training and test sets, fitting LDA on MFCCs from amplitude-modulated calls, then predicting bird IDs on tonal calls. This appears to evaluate different call types for known identities, not unseen-bird generalization (design interpretation).
Annotation structure: The 2020 table links annotation_ref values to Raven selections, but multiple selections may share one reference, so an annotation row is not necessarily one independent call. Its free-text behavior field describes behavior during or just before vocalization and may hold multiple values; 2020 sequence context is documented in a separate workbook. The 2021 table instead keys by file and selection and includes structured call type, context and sequence fields.
Rights and access: The article is listed CC BY 4.0; the pinned analysis repository is MIT; the separate Edmond v1 dataset record declares MIT. The 34,517-byte local CSV's Git blob SHA-1 matches the pinned GitHub copy, but it has not been matched to an Edmond archive member. Edmond's current docs require accepting both repository terms and the dataset license before download; ZIP-member access is documented but does not establish target-file presence or access. No Edmond file was downloaded and no terms were accepted.

Reading location: Final-article abstract and targeted Methods/Figure 4 passages surfaced by the primary PMC index; official pinned GitHub README at commit 293b2c8d0c3b8e21a4eaa0511fe563207095765a; local checksum receipt; official Edmond v1 API, Data Access & Usage guide, ZIP-viewer FAQ, Terms of Use and the documentation homepage's July 20, 2026 terms notice. Direct full-text retrieval was CAPTCHA/timeout limited; supplement, complete analysis workflow, Edmond archive member and raw audio were not reviewed.

What this does — and doesn’t — establish

The research target is individual identity, not behavior, context or semantics. Figure 4 describes splitting recordings and predicting marked bird IDs across call types; this appears to test transfer for known individuals, not generalization to unseen birds (an interpretation of the design). In 2020, annotation_ref links annotation rows to Raven selections, and the README says multiple selections can share one reference; rows therefore are not necessarily independent call-level labels. The free-text behavior field describes behavior during or just before vocalizations and can contain multiple values; sequence context is documented separately. The 2021 schema differs. No observer-reliability result was inspected. No Edmond archive/member or deposited audio was acquired. The Edmond API's dataset-level fileAccessRequest flag does not resolve any individual file's access status.

What we can build on

Use this work as a caller-identity reproduction candidate and a warning to match splits to the claim. It cannot validate Talk2Nature's planned context task. Keep identity, recording and session groups connected; a new-bird or behavior claim needs its own held-out groups and independent labels.

The next reading task

The 34,517-byte local CSV is checksum-verified against the pinned GitHub file, but its identity inside the Edmond v1 archives remains unverified. Edmond's current terms require acceptance of platform terms and the dataset license before download, and published deposits are not submitter-deletable except by exceptional deaccession. Do not accept terms or retrieve files automatically. Only if a qualified reviewer confirms a concrete scientific need, prepare a single-file privacy/provenance review and obtain Raviv's explicit approval before any terms acceptance or retrieval; do not fetch full archives. Keep the sample outside training and redistribution meanwhile. For a first context benchmark, seek independently observed, timestamp-linked behavior rather than repurposing caller IDs or vocal-event notes.

Go to the sources

Monk parakeet vocal signatures ↗
Smeele et al. (2023), final Royal Society Open Science article. Targeted abstract and methods/figure text reviewed from the primary PMC index; direct full-text retrieval was CAPTCHA-limited and the ANU repository PDF fetch timed out. The ANU record lists Open Access under CC BY 4.0.
ANU repository record listing open access and CC BY 4.0 ↗
Monk parakeet analysis repository ↗
Official repository pinned at commit 293b2c8d0c3b8e21a4eaa0511fe563207095765a. The 34,517-byte local 2020 annotation CSV's Git blob SHA-1 (badde532a9712071e5048e3d45a8e3de95c618dc) matches this pinned source file; its SHA-256 is 2c771d60dcc955e28f500a091c242d19fafc928798a069b01ab13554c6e3e2de.
Monk parakeet deposited dataset ↗
Edmond v1 API rechecked October 13, 2026: released version published 2023-09-07 declares MIT and lists three archives totaling 100,498,691,932 bytes with provider-supplied MD5 values. Version metadata reports fileAccessRequest=true, but no individual file's access state was resolved. No Edmond archive or member was downloaded. The local GitHub annotation-table copy is checksum-verified against its pinned GitHub commit, but has not been matched to a member inside the Edmond archives.
Edmond data access and reuse terms ↗
Current Edmond user guide: downloading requires acceptance of Edmond's Terms of Use and the dataset license. It documents direct single-file downloads and ZIP downloads, capped at 4.7 GB for the all-file or multi-file options.
Edmond ZIP preview and member access ↗
Official Edmond FAQ documents a ZIP viewer that lists files and permits individual-member downloads without first downloading the complete ZIP. It does not confirm that the monk-parakeet annotation CSV is present or accessible in this deposit.
Edmond Terms of Use (current repository terms) ↗
Official terms say published datasets are preserved for at least ten years and normally cannot be erased by submitters except through exceptional deaccession decided by Edmond. The registration section describes service use for Max Planck Society projects and partners; the page does not separately explain how that wording applies to external users downloading an existing public dataset.
Edmond documentation notice: new terms effective July 20, 2026 ↗
The official documentation homepage says its latest update was August 6, 2026 and that new Edmond Terms of Use and a curation process took effect July 20, 2026.

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.