We present an adaptation of the deep convolutional network Inception-v4 tailored to solving bioacoustic classification problems. Bird sound classification was treated as if it were an image classification problem by a transfer learning of Inception. Inception, the state-of-the-art in image classification, was used together with an attention algorithm, to (multiscale) time-frequency representations or images of bird sounds. This has resulted in an efficient pipeline, that we call Soundception. Soundception scored highest on all tasks in the BirdClef2017 challenge. It reached 0.714 Mean Average Precision in the task that asked for classification of 1500 bird species. To our knowledge Soundception is currently the most effective model for biodiversity monitoring of complex soundscapes.
[1]
Hervé Glotin,et al.
LifeCLEF 2017 Lab Overview: Multimedia Species Identification Challenges
,
2017,
CLEF.
[2]
Yoshua Bengio,et al.
Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
,
2015,
ICML.
[3]
Sergey Ioffe,et al.
Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning
,
2016,
AAAI.
[4]
Hervé Glotin,et al.
LifeCLEF Bird Identification Task 2016: The arrival of Deep learning
,
2016,
CLEF.
[5]
Thomas Hofmann,et al.
Audio Based Bird Species Identification using Deep Learning Techniques
,
2016,
CLEF.