Dataset Viewer
Auto-converted to Parquet
Search is not available for this dataset
audio
sequencelengths
268
5.78M
source_dataset
stringclasses
13 values
metadata
stringlengths
38
340
id
stringlengths
36
36
created_at
stringdate
2025-03-10 14:16:06
2025-03-10 14:16:06
license
stringclasses
14 values
file_name
stringlengths
6
170
instruction
stringclasses
13 values
instruction_text
stringclasses
13 values
output
stringlengths
3
145
task
stringclasses
3 values
dataset_name
stringclasses
22 values
[0.0007625818252563477,0.0008270144462585449,0.0008803009986877441,0.0009348392486572266,0.000986099(...TRUNCATED)
DCASE-2021-Task-5
{"duration": 2.0, "sample_rate": 16000}
27f7fdb0-9258-4615-b7fb-1edcdba163c0
2025-03-10T14:16:07.175000
CC-BY
y1.051_001.wav
"<Audio><AudioHere></Audio> Which of these, if any, are present in the audio recording? American Red(...TRUNCATED)
"Which of these, if any, are present in the audio recording? American Redstart, Bay-breasted Warbler(...TRUNCATED)
None
classification
dcase
[-0.02056884765625,-0.016876220703125,-0.012451171875,-0.0184326171875,-0.02276611328125,-0.02578735(...TRUNCATED)
iNaturalist
"{\"recordist\": \"Vasco Frederico\", \"url\": \"https://static.inaturalist.org/sounds/565419.wav?16(...TRUNCATED)
3cd99b4d-c685-4e3c-b650-6e630431517e
2025-03-10T14:16:07.517000
http://creativecommons.org/licenses/by-nc/4.0/
565419.flac
<Audio><AudioHere></Audio> Caption the audio, using the common name for any animal species.
Caption the audio, using the common name for any animal species.
European Serin
captioning
captioning
[0.0005804896354675293,0.0006128549575805664,0.0006529092788696289,0.0006754398345947266,0.000670194(...TRUNCATED)
DCASE-2021-Task-5
{"duration": 2.0, "sample_rate": 16000}
1c13ac2e-6a5f-4d3e-a1e6-6a80d12b2ecc
2025-03-10T14:16:07.170000
CC-BY
h1.052_024.wav
"<Audio><AudioHere></Audio> Which of these, if any, are present in the audio recording? American Red(...TRUNCATED)
"Which of these, if any, are present in the audio recording? American Redstart, Bay-breasted Warbler(...TRUNCATED)
None
classification
dcase
[0.0042266845703125,0.0056915283203125,0.0026397705078125,0.009521484375,0.0027618408203125,0.011703(...TRUNCATED)
Hainan Gibbons
{"duration": 4.0, "sample_rate": 9600}
4739371e-e957-4017-8f30-9458c8416afc
2025-03-10T14:16:07.010000
CC-BY-NC-SA
HGSM3SOL_0+1_20160320_054700.267_025.wav
"<Audio><AudioHere></Audio> Which of these, if any, are present in the audio recording? Single pulse(...TRUNCATED)
"Which of these, if any, are present in the audio recording? Single pulse gibbon call, Multiple puls(...TRUNCATED)
None
detection
gibbons
[0.022552490234375,0.024139404296875,0.018310546875,0.0193023681640625,0.0221405029296875,0.02369689(...TRUNCATED)
Hainan Gibbons
{"duration": 4.0, "sample_rate": 9600}
b5c10cce-944e-44ef-8ee7-d46d64eaa024
2025-03-10T14:16:06.913000
CC-BY-NC-SA
HGSM3C_0+1_20160501_051500.015_006.wav
"<Audio><AudioHere></Audio> Which of these, if any, are present in the audio recording? Single pulse(...TRUNCATED)
"Which of these, if any, are present in the audio recording? Single pulse gibbon call, Multiple puls(...TRUNCATED)
None
detection
gibbons
[0.0006712675094604492,0.0006725788116455078,0.0006719827651977539,0.0006719231605529785,0.000673413(...TRUNCATED)
DCASE-2021-Task-5
{"duration": 2.0, "sample_rate": 16000}
9d7aa3dc-b8df-48fb-9596-f9036ec0cf65
2025-03-10T14:16:07.170000
CC-BY
h1.052_018.wav
"<Audio><AudioHere></Audio> Which of these, if any, are present in the audio recording? American Red(...TRUNCATED)
"Which of these, if any, are present in the audio recording? American Redstart, Bay-breasted Warbler(...TRUNCATED)
None
classification
dcase
[0.0676422119140625,0.008148193359375,-0.066650390625,0.0600738525390625,0.1498565673828125,0.071899(...TRUNCATED)
Hainan Gibbons
{"duration": 4.0, "sample_rate": 9600}
f9fd8a1a-c644-4671-98b0-0959c697f113
2025-03-10T14:16:06.930000
CC-BY-NC-SA
HGSM3C_0+1_20160501_051500.231_011.wav
"<Audio><AudioHere></Audio> Which of these, if any, are present in the audio recording? Single pulse(...TRUNCATED)
"Which of these, if any, are present in the audio recording? Single pulse gibbon call, Multiple puls(...TRUNCATED)
None
detection
gibbons
[0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,-0.000030517578125,-0.000030517578125,-0.000030517578125,-0(...TRUNCATED)
Xeno-canto
"{\"recordist\": \"Marc Anderson\", \"url\": \"https://xeno-canto.org/sounds/uploaded/EHGWCIGILC/XC4(...TRUNCATED)
1b2740e9-75d1-4fcf-ba0b-3b2144657550
2025-03-10T14:16:07.276000
CC BY-NC
XC434556-Nightjar_Spotted_MX275_Carnarvon_DuskA_22ndAug18.flac
<Audio><AudioHere></Audio> What is the common name for the focal species in the audio?
What is the common name for the focal species in the audio?
Spotted Nightjar
classification
unseen-genus-cmn
[-0.001983642578125,-0.001678466796875,-0.002044677734375,-0.001800537109375,-0.00189208984375,-0.00(...TRUNCATED)
HumBugDB
{"duration": 61.38, "sample_rate": 44100}
96aeba30-eaad-44c7-8084-ffa28072a7bf
2025-03-10T14:16:06.715000
CC-BY
220372.wav
<Audio><AudioHere></Audio> What is the common name for the focal species in the audio?
What is the common name for the focal species in the audio?
ae aegypti
classification
humbugdb
[0.0,-0.000030517578125,-0.000030517578125,-0.000030517578125,-0.000030517578125,-0.000030517578125,(...TRUNCATED)
iNaturalist
"{\"recordist\": \"pisum\", \"url\": \"https://static.inaturalist.org/sounds/53430.wav?1577310957\",(...TRUNCATED)
af9cfc9d-4400-4c6e-93fb-b6da96411111
2025-03-10T14:16:07.348000
http://creativecommons.org/licenses/by-nc/4.0/
53430.flac
<Audio><AudioHere></Audio> Caption the audio, using the common name for any animal species.
Caption the audio, using the common name for any animal species.
The sound of a Yellow-rumped Warbler in a Chinese tallow tree.
captioning
captioning
End of preview. Expand in Data Studio

BEANS-Zero

Version: 0.1.0 Created on: 2025-04-12 Creators:

Overview

BEANS-Zero is a bioacoustics benchmark designed to evaluate multimodal audio-language models in zero-shot settings. Introduced in the paper NatureLM-audio paper (Robinson et al., 2025), it brings together tasks from both existing datasets and newly curated resources.

The benchmark focuses on models that take a bioacoustic audio input (e.g., bird or mammal vocalizations) and a text instruction (e.g., "What species is in this audio?"), and return a textual output (e.g., "Taeniopygia guttata"). As a zero-shot benchmark, BEANS-Zero contains only a test split—no training or in-context examples are provided.

Many tasks originate from the original BEANS benchmark, but BEANS-Zero adds new datasets and task types that broaden the evaluation scope.

Tasks and Applications

BEANS-Zero supports a wide range of zero-shot evaluation tasks, including:

  • Audio Classification — Identify species or sound categories from animal vocalizations.
  • Audio Detection — Detect the presence of species in long-form recordings.
  • Audio Captioning — Generate natural language descriptions of acoustic scenes.

Dataset Composition

BEANS-Zero combines data from several well-known sources. There are total of 91,965 samples (examples). It consists of two main groups:

Original BEANS Tasks

  • esc-50: Generic environmental sound classification with 50 labels (Piczak, 2015, License: CC-BY-NC)
  • watkins: Marine mammal species classification with 31 species (Sayigh et al., 2016, free for personal and academic use)
  • cbi: Bird species classification with 264 labels from the CornellBird Identification competition hosted on Kaggle (Howard et al., 2020, License: CC-BY-NC-SA)
  • humbugdb: Mosquito wingbeat sound classification into 14 species (Kiskin et al., 2021, License: CC-BY)
  • enabirds: Bird dawn chorus detection with 34 species (Chronister et al., 2021, License: CC0)
  • hiceas: Minke whale detection from the Hawaiian Islands Cetacean and Ecosystem Assessment Survey (HICEAS) (NOAA, 2022, free without restriction)
  • rfcx: Bird and frog detection from the Rainforest Connection(RFCx) data with 24 species (LeBien et al., 2020, usage allowed for academic research)
  • gibbons: Hainan gibbon detection with 3 call type labels (Dufourq et al., 2021, License: CC-BY-NC-SA)

Newly Added Subsets

  • unseen-species-*: Unseen species classification with 200 species held out from AnimalSpeak (Robinson et al., 2024), with each sub-dataset using common (cmn), scientific (sci), or taxonomic (tax) names
  • unseen-genus-*: Generalize to unseen genera (cmn/sci/tax)
  • unseen-family-*: Generalize to unseen families (cmn/sci/tax)
  • lifestage: Predicting the lifestage of birds across multiple species (e.g., adult, juvenile), curated from xeno-canto
  • call-type: Classifying song vs. call across multiple bird species, curated from xeno-canto
  • captioning: Captioning bioacoustic audio on AnimalSpeak (Robinson et al., 2024)
  • zf-indv: Determining whether a recording contains multiplezebra finches, using programmatically generated mixtures (1–4 individuals) (Elie and Theunissen, 2016)

Each sample is labeled with its source dataset and license.

Usage

import numpy as np
from datasets import load_dataset

ds = load_dataset("EarthSpeciesProject/BEANS-Zero", split="test") 

# see the contents at a glance
print(ds)
# get audio for the first sample in the dataset, the 0th index
audio = np.array(ds[0]["audio"])
print(audio.shape)

# get the instruction (prompt / query) for that sample
print(ds[0]["instruction_text"])
# the desired output (should *only* be used for evaluation)
print(ds[0]["output"])

# the component datasets of BEANS-Zero are:
components, dataset_sample_counts = np.unique(ds["dataset_name"], return_counts=True)

# if you want to select a subset of the data, e.g. 'esc50'
idx = np.where(np.array(ds["dataset_name"]) == "esc50")[0]
esc50 = ds.select(idx)
print(esc50)
# To stream the dataset instead of downloading it, first
ds = load_dataset("EarthSpeciesProject/BEANS-Zero", split="test", streaming=True)

for i, sample in enumerate(ds):
    # check one sample
    break
print(sample.keys())

Data Fields

The following fields are present in each example:

  • source_dataset (str): One of the source datasets mentioned above
  • audio (Sequence[float]): The audio data as a list of floats.
  • id (str): Sample uuid.
  • created_at (str): Sample creation datetime in UTC
  • metadata (str): Each sample can have a different duration (in seconds) and a different sample rate (in Hz). The 'metadata' is a JSON string containing these two fields.
  • file_name (str): Audio file name
  • instruction (str): A prompt (a query) corresponding to the audio for your audio-text model with a placeholder for audio tokens. E.g. ' What is the scientific name for the focal species in the audio?'
  • instruction_text (str): Same as instruction but without the placeholder for audio tokens.
  • output (str): The expected output from the model
  • task (str): The task type e.g. classification / detection / captioning.
  • dataset_name (str): Names corresponding to the evaluation tasks, e.g. 'esc50' or 'unseen-family-sci'.
  • license (str): The license of the dataset. For example, 'CC-BY-NC' or 'CC0'.

Licensing

Due to its composite nature, BEANS-Zero is subject to multiple licenses. Individual samples have the "license" field indicating the specific license for that sample. The dataset is not intended for commercial use, and users should adhere to the licenses of the individual datasets.

Citation

If you use BEANS-Zero, please cite the following:

@inproceedings{robinson2025naturelm,
  title     = {NatureLM-audio: an Audio-Language Foundation Model for Bioacoustics},
  author    = {David Robinson and Marius Miron and Masato Hagiwara and Olivier Pietquin},
  booktitle = {Proceedings of the International Conference on Learning Representations (ICLR)},
  year      = {2025},
  url       = {https://openreview.net/forum?id=hJVdwBpWjt}
}

Contact

For questions, comments, or contributions, please contact:

  • David Robinson (david at earthspecies dot org)
  • Marius Miron (marius at earthspecies dot org)
  • Masato Hagiwara (masato at earthspecies dot org)
  • Gagan Narula (gagan at earthspecies dot org)
  • Milad Alizadeh (milad at earthspecies dot org)
Downloads last month
29