Talk2NaturePlants & fungi / Recording study
Plant acoustic emissions: detection is not communication
Source year 2023 · Note reviewed 2026-10-13
Khait et al. report airborne ultrasonic emissions from stressed tomato and tobacco in an acoustic box and greenhouse, and classify plant condition from recordings. The system used paired CM16 ultrasound microphones per plant at 500 kHz, with a 20 kHz high-pass filter; an event counted as plant-associated only when both microphones aimed at that plant recorded it. They report leave-one-plant-out evaluation: a 1-hour greenhouse dry-versus-irrigated classification reached 84% balanced accuracy, and a two-range soil-moisture classification in a nine-day dry-down of 23 tomato plants reached 81%. These are emission and physiological-state results, not decoded messages or tested receiver responses.
What we reviewed · Full paper, critical context and component-rights metadata reviewed
Read the main 2023 Cell article and its full STAR Methods, including the recording, preprocessing, model evaluation and results sections; checked the relevant figure legends and selected supporting captions. Read the 2024 Opinion's historical framing, mechanism discussion, receiver-evidence conclusion and open questions. Also checked the exact Dryad record, Dryad's live CC0 policy and the linked Zenodo v1 classifier record. The Cell publisher endpoint was inaccessible, so its linked full-paper copy was used. No audio, archive, code or model was downloaded or run; no result was independently reproduced.
Not yet independently reviewed by a scientific specialist.
What this does — and doesn’t — establish
The experiments used specialized ultrasonic hardware and controlled acoustic-box/greenhouse settings; ordinary phone microphones are not an equivalent sensor. The plant-held-out folds reduce leakage from repeated sounds of the same plant, but do not establish generalization to new sites, devices, species or field noise. The paper's 3–5 m animal detection range is estimated from reported sound levels and published hearing thresholds, not measured animal response. Nardini et al.'s 2024 Opinion places acoustic emissions in a long history of plant physiology and discusses xylem cavitation and other physical mechanisms; it says ecological receiver use still needed evidence at that time. The later moth study in source 16 supplies a bounded receiver-response result in one moth/plant setting, not evidence of plant intention or a general language. No independent reproduction or qualified plant-acoustics review has been obtained. Dryad lists the recordings under its CC0 repository policy, but the archive contents were not inspected; the Zenodo classifier record has no license stated.
What we can build on
Treat this as a sensor-and-physiology research branch. Establish that a specialized sensor detects a bounded physical event, then test its context and any receiver response independently; do not turn a stress classifier into a conversation claim.
See the related tools and reading paths →
The next reading task
If a plant pilot is pursued, first obtain qualified plant-acoustics review and define a calibrated sensor question that ordinary phones cannot answer. Before any benchmark, inspect the exact Dryad archive and its labels and grouping, lock a plant-held-out evaluation, and resolve the separate Zenodo code license. Keep this branch separate from animal vocalization models and participant-data intake.
Go to the sources
- Plant acoustic emissions ↗
Khait et al. (2023), Cell primary study of airborne ultrasonic emissions and plant-condition classification. Full article and STAR Methods reviewed; the paper does not test receiver behavior or translation.
Full published paper copy and STAR Methods used for review ↗
- Moth responses to plant sounds ↗
Seltzer et al. Female moths incorporate plant acoustic emissions into their oviposition decision-making process. eLife; version of record 5 January 2026. The article is CC BY 4.0; earlier reviewed preprints appeared in 2024 and 2025.
- Nardini, Cochard and Mayr: Talk is cheap: rediscovering sounds made by plants ↗
2024 Trends in Plant Science Opinion. It places plant acoustic emissions in a long physiological research history, discusses cavitation and other passive physical mechanisms, and says receiver use needs experimental evidence. It is a critical synthesis, not a new experiment.
- Khait et al. plant recordings, Dryad dataset ↗
The exact Dryad page lists a 10 MB PlantSounds.zip and an 806-byte README. The README describes two-millisecond WAV clips sampled at 500 kHz and labels for dry/cut tomato and tobacco, empty pots and greenhouse noise.
- Dryad good data practices and CC0 dataset policy ↗
Dryad's official guidance says data files submitted to its repository must meet CC0 terms and be free of conflicting copyright or terms; software may be hosted separately under other licenses.
- Khait et al. CNN plant-sound classifier training function, Zenodo v1 ↗
The linked Zenodo software record describes a 2.4 kB binary CNN training function, published February 6, 2023. Its License field is blank.
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.