Deep Learning for Audio Event Detection and Tagging on Low-Resource Datasets

Morfi, Veronica; Stowell, Dan

doi:10.3390/app8081397

Cited by 41 publications

(38 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Desjonquères, Rybak, Ulloa, et al., ; Dutilleux & Curé, ) to more refined pattern recognition via cross‐correlation (Ulloa et al., ) as well as more complex methods using machine learning (e.g. Morfi & Stowell, ; Xie, Towsey, Zhang, & Roe, ; Zhang, Towsey, Zhang, & Roe, ). Automatic detection and classification methods usually first involve feature extraction and then classification of these features (Sharan & Moir, ).…”

Section: How To Undertake Pam In Freshwatermentioning

confidence: 99%

Passive acoustic monitoring as a potential tool to survey animal and ecosystem processes in freshwater environments

2019

View full text Add to dashboard Cite

Biodiversity in freshwater habitats is decreasing faster than in any other type of environment, mostly as a result of human activities. Monitoring these losses can help guide mitigation efforts. In most studies, sampling strategies predominantly rely on collecting animal and vegetal specimens. Although these techniques produce valuable data, they are invasive, time‐consuming and typically permit only limited spatial and temporal replication. There is need for the development of complementary methods. As observed in other ecosystems, freshwater environments host animals that emit sounds, either to communicate or as a by‐product of their activity. The main freshwater soniferous groups are amphibians, fish, and macroinvertebrates (mainly Coleoptera and Hemiptera, but also some Decapoda, Odonata, and Trichoptera). Biophysical processes such as flow or sediment transport also produce sounds, as well as human activities within aquatic ecosystems. Such animals and processes can be recorded, remotely and autonomously, and provide information on local diversity and ecosystem health. Passive acoustic monitoring (PAM) is an emerging method already deployed in terrestrial environments that uses sounds to survey environments. Key advantages of PAM are its non‐invasive nature, as well as its ability to record autonomously and over long timescales. All these research topics are the main aims of ecoacoustics, a new scientific discipline investigating the ecological role of sounds. In this paper, we review the sources of sounds present in freshwater environments. We then underline areas of research in which PAM may be helpful emphasising the role of PAM for the development of ecoacoustics. Finally, we present methods used to record and analyse sounds in those environments. Passive acoustics represents a potentially revolutionary development in freshwater ecology, enabling continuous monitoring of dynamic bio‐physical processes to inform conservation practitioners and managers.

show abstract

Section: How To Undertake Pam In Freshwatermentioning

confidence: 99%

Passive acoustic monitoring as a potential tool to survey animal and ecosystem processes in freshwater environments

2019

View full text Add to dashboard Cite

show abstract

“…A few examples of the NIPS4Bplus dataset and temporal annotations being used can be found in (Morfi and Stowell, 2018a) and (Morfi and Stowell, 2018b). First, in (Morfi and Stowell, 2018a), we use NIPS4Bplus to carry out the training and evaluation of a newly proposed multi-instance learning (MIL) loss function for audio event detection.…”

Section: Example Uses Of Nips4bplusmentioning

confidence: 99%

Peer Review #1 of "NIPS4Bplus: a richly annotated birdsong audio dataset (v0.2)"

2019

View full text Add to dashboard Cite

Recent advances in birdsong detection and classification have approached a limit due to the lack of fully annotated recordings. In this paper, we present NIPS4Bplus, the first richly annotated birdsong audio dataset, that is comprised of recordings containing bird vocalisations along with their active species tags plus the temporal annotations acquired for them. Statistical information about the recordings, their species specific tags and their temporal annotations are presented along with example uses. NIPS4Bplus could be used in various ecoacoustic tasks, such as training models for bird population monitoring, species classification, birdsong vocalisation detection and classification.

show abstract

“…Usually, bird vocalizations are segmented to improve the performance of the classifier. However, these segmentation algorithms are commonly too simple for real conditions in the field or follow a supervised learning scheme were a lot of manual work has to be done to label the vocalizations used for training [4].…”

Section: Introductionmentioning

confidence: 99%

Entendiendo el Desempeño Variable en el Marco de Trabajo MIL Profundo para la Detección Acústica de Aves Tropicales

Castro

Vargas-Masís

Alfaro-Rojas

2020

View full text Add to dashboard Cite

Se han propuesto muchos algoritmos de detección de audio para monitorear aves usando sus vocalizaciones. Entre estos algoritmos, las técnicas basadas en el aprendizaje profundo han tomado la delantera en términos de rendimiento a gran escala. Sin embargo, usualmente se requiere de mucho trabajo manual para etiquetar correctamente las vocalizaciones de aves en grandes conjuntos de datos. Una forma de abordar esta limitación es usar el marco de trabajo de aprendizaje de instancias múltiples (MIL), que modela cada grabación como una bolsa de instancias, es decir, una colección de segmentos de audio que se asocia con una etiqueta positiva si un pájaro está presente en la grabación. En este trabajo, modificamos una red profunda MIL propuesta previamente, para predecir la presencia o ausencia de aves en grabaciones de campo de un minuto. Exploramos el comportamiento y el rendimiento de la red cuando utilizamos un número diferente de coeficientes cepstrales de frecuencia de mel (MFCC) para representar las grabaciones. La mejor configuración encontrada logró un valor F de 0.77 sobre el conjunto de datos de validación.

show abstract

Deep Learning for Audio Event Detection and Tagging on Low-Resource Datasets

Cited by 41 publications

References 25 publications

Passive acoustic monitoring as a potential tool to survey animal and ecosystem processes in freshwater environments

Passive acoustic monitoring as a potential tool to survey animal and ecosystem processes in freshwater environments

Peer Review #1 of "NIPS4Bplus: a richly annotated birdsong audio dataset (v0.2)"

Entendiendo el Desempeño Variable en el Marco de Trabajo MIL Profundo para la Detección Acústica de Aves Tropicales

Contact Info

Product

Resources

About