On Multitask Loss Function for Audio Event Detection and Localization
This work addresses a specific bottleneck in audio event detection and localization for applications like surveillance or robotics, but it is incremental as it improves upon existing multitask approaches.
The paper tackles the problem of audio event localization and detection (SELD) by proposing a multitask regression model that uses a homogeneous mean squared error loss for both tasks, instead of the common heterogeneous loss functions. The result is a reduction in overall SELD error by approximately 10% absolute, outperforming the baseline on DCASE 2020 datasets.
Audio event localization and detection (SELD) have been commonly tackled using multitask models. Such a model usually consists of a multi-label event classification branch with sigmoid cross-entropy loss for event activity detection and a regression branch with mean squared error loss for direction-of-arrival estimation. In this work, we propose a multitask regression model, in which both (multi-label) event detection and localization are formulated as regression problems and use the mean squared error loss homogeneously for model training. We show that the common combination of heterogeneous loss functions causes the network to underfit the data whereas the homogeneous mean squared error loss leads to better convergence and performance. Experiments on the development and validation sets of the DCASE 2020 SELD task demonstrate that the proposed system also outperforms the DCASE 2020 SELD baseline across all the detection and localization metrics, reducing the overall SELD error (the combined metric) by approximately 10% absolute.