Query-by-Example Keyword Spotting system using Multi-head Attention and Softtriple Loss
This work addresses keyword spotting for user-defined queries, but it is incremental as it builds on existing methods with hybrid components.
The paper tackles the query-by-example keyword spotting task by proposing a neural network with multi-head attention and softtriple loss, achieving solid performance on internal and public datasets compared to a baseline.
This paper proposes a neural network architecture for tackling the query-by-example user-defined keyword spotting task. A multi-head attention module is added on top of a multi-layered GRU for effective feature extraction, and a normalized multi-head attention module is proposed for feature aggregation. We also adopt the softtriple loss - a combination of triplet loss and softmax loss - and showcase its effectiveness. We demonstrate the performance of our model on internal datasets with different languages and the public Hey-Snips dataset. We compare the performance of our model to a baseline system and conduct an ablation study to show the benefit of each component in our architecture. The proposed work shows solid performance while preserving simplicity.