1Faculty of Environment, Science and Economy, Department of Earth and Environmental Science, Centre for Geography and Environmental Science, University of Exeter, Penryn, TR10 9FE, UK2Yayasan Borneo Nature Indonesia, Palangka Raya, Kalimantan Tengah, 73112, Indonesia3Fakultas Kehutanan dan Pertanian, Universitas Muhammadiyah Palangkaraya, Palangka Raya, Kalimantan Tengah, 73111, Indonesia4K. Lisa Yang Center for Conservation Bioacoustics, Cornell Lab of Ornithology, Cornell University, Ithaca, NY 14850, USACorresponding author: A. F. Owens ([email protected])IntroductionThis collection contains 1,611 sound recordings of Bornean white-bearded gibbon (Hylobates albibarbis) “great call” vocalisations, split into training (1,089 recordings) and testing (522 recordings) datasets. These were used in Owens et al. (2024) to train a neural network with the purpose of automatically detecting great-call vocalisations in acoustic data. The recordings derive from the Mungku Baru Education and Research Forest (MBERF), a ~50 km2 area of tropical rainforest in Central Kalimantan, Indonesia (Buckley et al. 2019).Data collectionEight autonomous recording units (ARUs) (Song Meter SM4, Wildlife Acoustics, Maynard, Massachusetts) were deployed by WME and EE from 18th July 2018. These were strapped to trees, 5 meters above the ground, in a dispersed grid with approximately 1,200 meters between each device. The MBERF contains a mosaic of different forest types, and the location of the ARUs was intended to reflect this heterogeneity, with three situated in kerangas heath forest, three in low pole peat swamp forest, and two in mixed swamp forest (see Appendix, Table 1).The ARUs were set to record continuously from 4 am to 6 pm local time (WIB/UTC +7) daily to capture the full pre-dawn and diurnal period of ape calling. Default gain settings (microphone = 16 dB, preamplifier = 26 dB), stereo recording, and a sampling rate of 24 kHz were used for all recordings. Audio was captured in 16-bit Waveform Audio File Format (WAV) and saved as one-hour files. Files were automatically saved to the SD card as “ARU ID_YYYYMMDD_HHMMSS.wav”.To create training and testing datasets for the development of an automated detector, we used randomly stratified sampling by habitat and month to identify a subset of sound files for manual annotation. To do this, we selected recordings between 4-10 am from a single day (i.e., the same date for all recorders) from a randomly selected device for each habitat and repeated this selection at four-week intervals. The resultant subset contained 300 hours of recordings, covering 50 days spanning from October 2018 to December 2019. This included 96 hours of audio from heath forest, and 102 hours each from both low pole and mixed swamp forest habitats.Manual annotationThe selected sound files were opened in Raven Pro 1.6 (K. Lisa Yang Center for Conservation Bioacoustics, Ithaca, NY, USA) and visualised as spectrograms. These were generated using a 3462-sample Hann window with a 90% overlap and a 4096-sample Discrete Fourier Transformation. With assistance from a team of undergraduate interns (see Acknowledgments), each recording was listened to in full and visually scanned to identify great-call events. A selection was created for each event by drawing a box that fully captured the fundamental frequency of the call in the spectrogram, providing information about its time-frequency boundaries (Appendix, Figure 1).Each selection was annotated for its completeness and quality. A selection was marked “GC” (great call, i.e., clear) when the entirety of the call could be heard and was visually clear in the spectrogram, “F” (faint) when the whole call could be heard but was not fully shown in the spectrogram (and vice versa), and “VF” (very faint) when the whole call could neither be seen nor heard in full. In some cases, great calls were cut off by the beginning or the end of a recording and were annotated as “ER” (end of recording). These ER calls were not included in the training or testing datasets.Each selection was reviewed and edited where needed by AFO to reduce the influence of inter-observer variability. In total, 1,611 great calls were annotated: 839 “clear”, 256 “faint” and 516 “very faint” calls. These were then randomly split into training (210 hours, 1,089 calls) and testing (90 hours, 522 calls) datasets (see Appendix, Table 2).For this dataset, the annotated great calls were saved as clips and renamed as “ID_soundfile_annotation.wav” (Appendix, Figure 2). In this case, “ID” serves as a unique identifier for each great call recording.AcknowledgementsWe thank Universitas Muhammadiyah Palangkaraya and the Borneo Nature Foundation for access to and support in the MBERF and its research facilities. We thank the Indonesian government for permission to carry out this research (RISTEK-DIKTI permit #189/SIP/FRP/E5/Dit.KI/VI/2018 and 43/E5/E5.4/SIP.EXT/2019 issued to Wendy M. Erb). We thank Rido for his invaluable support in deploying and maintaining the ARUs and data in the field. We also thank Georgia Allen, Amy Barron, Sophie Carpenter, and Elena Gough for their contributions in manually annotating the dataset. In addition, we acknowledge the contributions of the Indigenous Dayak Ngaju community of Mungku Baru, including Pak Edo, Pak Yuli, Pak Viktor and Rico, who supported the data collection and shared their knowledge. Please note the sponsors of this research: Primate Conservation, Incorporated; British Academy; Conservation International; American Association of Physical Anthropologists; International Primatological Society; and American Institute for Indonesian Studies.ReferencesBuckley, B. J. W., Capilla, B. R., Maimunah, S., Adul, Armadyanto, Boyd, N., Cheyne, S. M., Iwan, Husson, S. J., Santiano, Salahudin, Ferisa, A., Namaskari, N., van Veen, F. J. F., Harrison, M. E. 2019. Biodiversity, forest structure & conservation importance of the Mungku Baru Education Forest, Rungan, Central Kalimantan, Indonesia. BNF Reports.Owens, A. F., Hockings, K. J., Imron, M. A., Madhusudhana, S., Mariaty, Setia, T. M., Sharma, M., Maimunah, S., Van Veen, F. J. F., Erb, W. M. 2024. Automated detection of Bornean white-bearded gibbon (Hylobates albibarbis) vocalizations using an open-source framework for deep learning. J. Acoust. Soc. Am. 156 (3): 1623–1632. https://doi.org/10.1121/10.0028268

1Faculty of Environment, Science and Economy, Department of Earth and Environmental Science, Centre for Geography and Environmental Science, University of Exeter, Penryn, TR10 9FE, UK2Yayasan Borneo Nature Indonesia, Palangka Raya, Kalimantan Tengah, 73112, Indonesia3Fakultas Kehutanan dan Pertanian, Universitas Muhammadiyah Palangkaraya, Palangka Raya, Kalimantan Tengah, 73111, Indonesia4K. Lisa Yang Center for Conservation Bioacoustics, Cornell Lab of Ornithology, Cornell University, Ithaca, NY 14850, USACorresponding author: A. F. Owens ([email protected])IntroductionThis collection contains 1,611 sound recordings of Bornean white-bearded gibbon (Hylobates albibarbis) “great call” vocalisations, split into training (1,089 recordings) and testing (522 recordings) datasets. These were used in Owens et al. (2024) to train a neural network with the purpose of automatically detecting great-call vocalisations in acoustic data. The recordings derive from the Mungku Baru Education and Research Forest (MBERF), a ~50 km2 area of tropical rainforest in Central Kalimantan, Indonesia (Buckley et al. 2019).Data collectionEight autonomous recording units (ARUs) (Song Meter SM4, Wildlife Acoustics, Maynard, Massachusetts) were deployed by WME and EE from 18th July 2018. These were strapped to trees, 5 meters above the ground, in a dispersed grid with approximately 1,200 meters between each device. The MBERF contains a mosaic of different forest types, and the location of the ARUs was intended to reflect this heterogeneity, with three situated in kerangas heath forest, three in low pole peat swamp forest, and two in mixed swamp forest (see Appendix, Table 1).The ARUs were set to record continuously from 4 am to 6 pm local time (WIB/UTC +7) daily to capture the full pre-dawn and diurnal period of ape calling. Default gain settings (microphone = 16 dB, preamplifier = 26 dB), stereo recording, and a sampling rate of 24 kHz were used for all recordings. Audio was captured in 16-bit Waveform Audio File Format (WAV) and saved as one-hour files. Files were automatically saved to the SD card as “ARU ID_YYYYMMDD_HHMMSS.wav”.To create training and testing datasets for the development of an automated detector, we used randomly stratified sampling by habitat and month to identify a subset of sound files for manual annotation. To do this, we selected recordings between 4-10 am from a single day (i.e., the same date for all recorders) from a randomly selected device for each habitat and repeated this selection at four-week intervals. The resultant subset contained 300 hours of recordings, covering 50 days spanning from October 2018 to December 2019. This included 96 hours of audio from heath forest, and 102 hours each from both low pole and mixed swamp forest habitats.Manual annotationThe selected sound files were opened in Raven Pro 1.6 (K. Lisa Yang Center for Conservation Bioacoustics, Ithaca, NY, USA) and visualised as spectrograms. These were generated using a 3462-sample Hann window with a 90% overlap and a 4096-sample Discrete Fourier Transformation. With assistance from a team of undergraduate interns (see Acknowledgments), each recording was listened to in full and visually scanned to identify great-call events. A selection was created for each event by drawing a box that fully captured the fundamental frequency of the call in the spectrogram, providing information about its time-frequency boundaries (Appendix, Figure 1).Each selection was annotated for its completeness and quality. A selection was marked “GC” (great call, i.e., clear) when the entirety of the call could be heard and was visually clear in the spectrogram, “F” (faint) when the whole call could be heard but was not fully shown in the spectrogram (and vice versa), and “VF” (very faint) when the whole call could neither be seen nor heard in full. In some cases, great calls were cut off by the beginning or the end of a recording and were annotated as “ER” (end of recording). These ER calls were not included in the training or testing datasets.Each selection was reviewed and edited where needed by AFO to reduce the influence of inter-observer variability. In total, 1,611 great calls were annotated: 839 “clear”, 256 “faint” and 516 “very faint” calls. These were then randomly split into training (210 hours, 1,089 calls) and testing (90 hours, 522 calls) datasets (see Appendix, Table 2).For this dataset, the annotated great calls were saved as clips and renamed as “ID_soundfile_annotation.wav” (Appendix, Figure 2). In this case, “ID” serves as a unique identifier for each great call recording.AcknowledgementsWe thank Universitas Muhammadiyah Palangkaraya and the Borneo Nature Foundation for access to and support in the MBERF and its research facilities. We thank the Indonesian government for permission to carry out this research (RISTEK-DIKTI permit #189/SIP/FRP/E5/Dit.KI/VI/2018 and 43/E5/E5.4/SIP.EXT/2019 issued to Wendy M. Erb). We thank Rido for his invaluable support in deploying and maintaining the ARUs and data in the field. We also thank Georgia Allen, Amy Barron, Sophie Carpenter, and Elena Gough for their contributions in manually annotating the dataset. In addition, we acknowledge the contributions of the Indigenous Dayak Ngaju community of Mungku Baru, including Pak Edo, Pak Yuli, Pak Viktor and Rico, who supported the data collection and shared their knowledge. Please note the sponsors of this research: Primate Conservation, Incorporated; British Academy; Conservation International; American Association of Physical Anthropologists; International Primatological Society; and American Institute for Indonesian Studies.ReferencesBuckley, B. J. W., Capilla, B. R., Maimunah, S., Adul, Armadyanto, Boyd, N., Cheyne, S. M., Iwan, Husson, S. J., Santiano, Salahudin, Ferisa, A., Namaskari, N., van Veen, F. J. F., Harrison, M. E. 2019. Biodiversity, forest structure & conservation importance of the Mungku Baru Education Forest, Rungan, Central Kalimantan, Indonesia. BNF Reports.Owens, A. F., Hockings, K. J., Imron, M. A., Madhusudhana, S., Mariaty, Setia, T. M., Sharma, M., Maimunah, S., Van Veen, F. J. F., Erb, W. M. 2024. Automated detection of Bornean white-bearded gibbon (Hylobates albibarbis) vocalizations using an open-source framework for deep learning. J. Acoust. Soc. Am. 156 (3): 1623–1632. https://doi.org/10.1121/10.0028268

1Faculty of Environment, Science and Economy, Department of Earth and Environmental Science, Centre for Geography and Environmental Science, University of Exeter, Penryn, TR10 9FE, UK2Yayasan Borneo Nature Indonesia, Palangka Raya, Kalimantan Tengah, 73112, Indonesia3Fakultas Kehutanan dan Pertanian, Universitas Muhammadiyah Palangkaraya, Palangka Raya, Kalimantan Tengah, 73111, Indonesia4K. Lisa Yang Center for Conservation Bioacoustics, Cornell Lab of Ornithology, Cornell University, Ithaca, NY 14850, USACorresponding author: A. F. Owens ([email protected])IntroductionThis collection contains 1,611 sound recordings of Bornean white-bearded gibbon (Hylobates albibarbis) “great call” vocalisations, split into training (1,089 recordings) and testing (522 recordings) datasets. These were used in Owens et al. (2024) to train a neural network with the purpose of automatically detecting great-call vocalisations in acoustic data. The recordings derive from the Mungku Baru Education and Research Forest (MBERF), a ~50 km2 area of tropical rainforest in Central Kalimantan, Indonesia (Buckley et al. 2019).Data collectionEight autonomous recording units (ARUs) (Song Meter SM4, Wildlife Acoustics, Maynard, Massachusetts) were deployed by WME and EE from 18th July 2018. These were strapped to trees, 5 meters above the ground, in a dispersed grid with approximately 1,200 meters between each device. The MBERF contains a mosaic of different forest types, and the location of the ARUs was intended to reflect this heterogeneity, with three situated in kerangas heath forest, three in low pole peat swamp forest, and two in mixed swamp forest (see Appendix, Table 1).The ARUs were set to record continuously from 4 am to 6 pm local time (WIB/UTC +7) daily to capture the full pre-dawn and diurnal period of ape calling. Default gain settings (microphone = 16 dB, preamplifier = 26 dB), stereo recording, and a sampling rate of 24 kHz were used for all recordings. Audio was captured in 16-bit Waveform Audio File Format (WAV) and saved as one-hour files. Files were automatically saved to the SD card as “ARU ID_YYYYMMDD_HHMMSS.wav”.To create training and testing datasets for the development of an automated detector, we used randomly stratified sampling by habitat and month to identify a subset of sound files for manual annotation. To do this, we selected recordings between 4-10 am from a single day (i.e., the same date for all recorders) from a randomly selected device for each habitat and repeated this selection at four-week intervals. The resultant subset contained 300 hours of recordings, covering 50 days spanning from October 2018 to December 2019. This included 96 hours of audio from heath forest, and 102 hours each from both low pole and mixed swamp forest habitats.Manual annotationThe selected sound files were opened in Raven Pro 1.6 (K. Lisa Yang Center for Conservation Bioacoustics, Ithaca, NY, USA) and visualised as spectrograms. These were generated using a 3462-sample Hann window with a 90% overlap and a 4096-sample Discrete Fourier Transformation. With assistance from a team of undergraduate interns (see Acknowledgments), each recording was listened to in full and visually scanned to identify great-call events. A selection was created for each event by drawing a box that fully captured the fundamental frequency of the call in the spectrogram, providing information about its time-frequency boundaries (Appendix, Figure 1).Each selection was annotated for its completeness and quality. A selection was marked “GC” (great call, i.e., clear) when the entirety of the call could be heard and was visually clear in the spectrogram, “F” (faint) when the whole call could be heard but was not fully shown in the spectrogram (and vice versa), and “VF” (very faint) when the whole call could neither be seen nor heard in full. In some cases, great calls were cut off by the beginning or the end of a recording and were annotated as “ER” (end of recording). These ER calls were not included in the training or testing datasets.Each selection was reviewed and edited where needed by AFO to reduce the influence of inter-observer variability. In total, 1,611 great calls were annotated: 839 “clear”, 256 “faint” and 516 “very faint” calls. These were then randomly split into training (210 hours, 1,089 calls) and testing (90 hours, 522 calls) datasets (see Appendix, Table 2).For this dataset, the annotated great calls were saved as clips and renamed as “ID_soundfile_annotation.wav” (Appendix, Figure 2). In this case, “ID” serves as a unique identifier for each great call recording.AcknowledgementsWe thank Universitas Muhammadiyah Palangkaraya and the Borneo Nature Foundation for access to and support in the MBERF and its research facilities. We thank the Indonesian government for permission to carry out this research (RISTEK-DIKTI permit #189/SIP/FRP/E5/Dit.KI/VI/2018 and 43/E5/E5.4/SIP.EXT/2019 issued to Wendy M. Erb). We thank Rido for his invaluable support in deploying and maintaining the ARUs and data in the field. We also thank Georgia Allen, Amy Barron, Sophie Carpenter, and Elena Gough for their contributions in manually annotating the dataset. In addition, we acknowledge the contributions of the Indigenous Dayak Ngaju community of Mungku Baru, including Pak Edo, Pak Yuli, Pak Viktor and Rico, who supported the data collection and shared their knowledge. Please note the sponsors of this research: Primate Conservation, Incorporated; British Academy; Conservation International; American Association of Physical Anthropologists; International Primatological Society; and American Institute for Indonesian Studies.ReferencesBuckley, B. J. W., Capilla, B. R., Maimunah, S., Adul, Armadyanto, Boyd, N., Cheyne, S. M., Iwan, Husson, S. J., Santiano, Salahudin, Ferisa, A., Namaskari, N., van Veen, F. J. F., Harrison, M. E. 2019. Biodiversity, forest structure & conservation importance of the Mungku Baru Education Forest, Rungan, Central Kalimantan, Indonesia. BNF Reports.Owens, A. F., Hockings, K. J., Imron, M. A., Madhusudhana, S., Mariaty, Setia, T. M., Sharma, M., Maimunah, S., Van Veen, F. J. F., Erb, W. M. 2024. Automated detection of Bornean white-bearded gibbon (Hylobates albibarbis) vocalizations using an open-source framework for deep learning. J. Acoust. Soc. Am. 156 (3): 1623–1632. https://doi.org/10.1121/10.0028268