Published September 10, 2020 | Version 1.0.0
Dataset Open

Automated detection of Hainan gibbon calls for passive acoustic monitoring

  • 1. African Institute for Mathematical Sciences; Stellenbosch University
  • 2. Centre for Research into Ecological and Environmental Modelling, University of St Andrews; Centre for Statistics in Ecology, the Environment, and Conservation, University of Cape Town; African Institute for Mathematical Sciences
  • 3. Institute of Zoology, Zoological Society of London; Department of Biological Sciences, Northern Illinois University
  • 4. School of Biological Sciences, University of Utah
  • 5. Institute of Zoology, Zoological Society of London
  • 6. Department of Life Sciences, University of Roehampton
  • 7. Living Collections, Zoological Society of London
  • 8. Bawangling National Nature Reserve

Description

Data accompanying the paper: "Automated detection of Hainan gibbon calls for passive acoustic monitoring"

Please cite this dataset as:

Dufourq, Emmanuel and Durbach, Ian and Hansford, James and Hoepfner, Amanda and Ma, Heidi and Bryant, Jessica and Stender, Christina and Li, Wenyong and Liu, Zhiwei and Chen, Qing and Zhou, Zhaoli and Turvey, Samuel. (2020). Automated detection of Hainan gibbon calls for passive acoustic monitoring. BioRxiv doi: https://doi.org/10.1101/2020.09.07.285502

The Hainan gibbon is the world's rarest primate and one of the world's rarest mammals, with only a single population of about 30 individuals surviving in Bawangling National Nature Reserve (BNNR), Hainan, China. Eight Song Meter SM3 recorders (Wildlife Acoustics, Maynard, Massachusetts) were used to collect acoustic data from 1 March to 20 August 2016 within BNNR. Recorders were attached to trees at approximately 1.5 meters from the ground in tropical evergreen forest. Recorders were set to record for eight hours each day from the time of sunrise, which varied between approximately 05:00 and 06:00 during the study period. Devices did not record continuously throughout the entire survey period due to logistical and technical issues; in total, survey days per recorder varied between 79 and 129 days, and roughly 6,000 hours of recordings were collected. The majority of recordings were made with a sampling rate of 9,600Hz and bit depth of 16, with isolated recordings at 28,800Hz.

We provide the audio data (.wav) used to train and test our neural network classifier along with the corresponding labelled text files (.data).

Files provided

Train.zip - contains the training .wav audio files

Train_Labels.zip - contains the labels for the training data

Test.zip - contains the testing .wav audio files

Test_Labels.zip - contains the labels for the test data

Extra_Labelled_Data.zip - contains extra data that was labelled and non-gibbon calls used for training

Extra_Labels.zip - contains the labels for the extra labelled data

Unlabelled_Data.zip - contains additional .wav audio files which have not been labelled. These are split into various files (1-15) and can be downloaded individually.

Code.zip - contains all the software scripts and notebooks

Manual-zip - contains the user manual

Labels

The names of the labelled files start with either "g_" or "n_", for example "g_HGSM3D_0+1_20160429_051600.data" and "n_HGSM3D_0+1_20160429_051600.data". Files starting with "g_" contain the timestamps of the gibbon calls, and files starting with "n_" contain the timestamps of non-gibbon calls (e.g. background noise and bird calls). An audio file will have both a "g_" and "n_" file. Each file has the following format: Start,End,Duration,Type,Notes, where "start" denotes the start time in seconds, "end" denotes the end time in seconds, "duration" denotes the duration (end - start) in seconds, "type" denotes the type of call/noise and "notes" are additional notes which we labelled.

Types

The legend for the "type" column in the labelled files is defined as follows. The types for gibbon and non-gibbon files are different and we distinguish this below.

 

Gibbon files ("g_")

type 1 = one pulse gibbon call

type 2 = multiple pulse gibbon call (check "notes column" below)

type 3 = duet gibbon call

 

Notes column (only available in gibbon files)

One of the following: 2 pulse call, 3 pulse call, 4 pulse call, 5 pulse call or 6 pulse call.

 

Non-gibbon files ("n_")

type 1 = rain

type 2 = other species (e.g. birds)

type 3 = rain and other species

type 4 = rain and external noise (e.g. aircraft)

type 5 = natural sounds and external noise

type 6 = natural sounds and other species

Training files (containing gibbon calls):

HGSM3AC_0+1_20160309_055600
HGSM3AC_0+1_20160312_055400
HGSM3A_0+1_20160304_060000
HGSM3BD_0+1_20160305_060000
HGSM3AC_0+1_20160314_055200
HGSM3B_0+1_20150616_050500
HGSM3BD_0+1_20160402_053600
HGSM3D_0+1_20160429_051600
HGSM3B_0+1_20160305_060000
HGSM3C_0+1_20160501_051500
HGSM3SOL_0+1_20160320_054700
HGSM3SOL_0+1_20160405_053400
HGSM3BD_0+1_20160401_053700

Testing files:

HGSM3B_0+1_20160323_054500
HGSM3B_0+1_20160321_054700
HGSM3B_0+1_20160306_055900
HGSM3B_0+1_20160308_055700
HGSM3B_0+1_20160309_055600
HGSM3B_0+1_20160316_055100
HGSM3B_0+1_20160311_055500
HGSM3B_0+1_20160304_060000
HGSM3B_0+1_20160322_054600

Notes

We thank the Management Office of Bawangling National Nature Reserve for logistical assistance in the field. Fieldwork was funded by an Arcus Foundation grant to STT and a Wildlife Acoustics grant to JVB. ID is supported in part by funding from the National Research Foundation of South Africa (Grant ID 90782, 105782). ED is supported by a postdoctoral fellowship from the African Institute for Mathematical Sciences South Africa, Stellenbosch University and the Next Einstein Initiative. This work was carried out with the aid of a grant from the International Development Research Centre, Ottawa, Canada (www.idrc.ca), and with financial support from the Government of Canada, provided through Global Affairs Canada (GAC; www.international.gc.ca). We also thank the following rangers who contributed to data collection: Guang Wei, Zhong Zhao, Qing Lin, Jinbing Zhang, Zhicheng Zhang, Quanjin Li, Xiaoliang Fu, Zhengchong Zhou, Lubiao Huang, Zhengkun Ye, Zhenghai Zou, Jinqiang Wang, Wentao Han and Zengnan Xie.

Files

Code.zip

Files (159.3 GB)

Name Size Download all
md5:69d6f8f5e885104ea2d4b09f02389400
1.7 MB Preview Download
md5:38c175f62d9e81afcf8d10db64381052
6.0 GB Preview Download
md5:e3ee436f592e9ca12e440b6abd465660
5.6 kB Preview Download
md5:f441dfba13988835217aebe7c8df2508
334.3 kB Preview Download
md5:fe3cacb2a9c03ac881a21e601e104a99
4.3 kB Preview Download
md5:26d7cc245e6b56225b91587a592e06cf
11.1 GB Preview Download
md5:e712fa83e5cce97a6730b44f948f37c9
3.8 kB Preview Download
md5:8aaceee3f97ce0fdfb7de7e16c33ed88
11.9 GB Preview Download
md5:68ab4a57f1acdcb05f99c0fc6b815a72
18.7 kB Preview Download
md5:97ccc9c791cd7cb5298c47fba5722689
8.5 GB Preview Download
md5:1917331cb73de818a3351fabe14a7471
8.9 GB Preview Download
md5:0c9ea73057da6cfc61f10a2bef97ef0c
8.0 GB Preview Download
md5:511027a4e328bd7e9db2e0c2dc55e88c
11.1 GB Preview Download
md5:01ecbb46d6f5a5b727a75165859a011f
10.3 GB Preview Download
md5:aa0e10afaaaaba5ded9f6637d72a5ceb
10.3 GB Preview Download
md5:f3cb498e89596d495703e6776be9e022
4.6 GB Preview Download
md5:5f1758d5e0dc5109dc0d9ffaacb16795
8.6 GB Preview Download
md5:105dd3e9326643dce91254f741f299c0
8.3 GB Preview Download
md5:bcf2138ad34b1e70e2b5c3e832f8a649
8.0 GB Preview Download
md5:ba57196d226b039ff9bf6e69b108ec7e
8.5 GB Preview Download
md5:f733c86dd5d28d1f0ef3432d33d2c946
7.8 GB Preview Download
md5:d37f838d1c1acfecf612131728a3599b
9.1 GB Preview Download
md5:82affc8ea1c21a10badf1bed0cdee848
9.2 GB Preview Download
md5:51c460a86561cbb095468fe5a3752f9b
9.2 GB Preview Download