Multitaper mel-spectrograms for keyword spotting
This addresses the sensitivity of keyword spotting to feature quality, though it is incremental as it builds on existing techniques.
The paper tackled the problem of improving keyword spotting (KWS) by focusing on feature extraction, specifically using the multitaper technique to create enhanced features, and experimental results confirmed the advantages of this approach.
Keyword spotting (KWS) is one of the speech recognition tasks most sensitive to the quality of the feature representation. However, the research on KWS has traditionally focused on new model topologies, putting little emphasis on other aspects like feature extraction. This paper investigates the use of the multitaper technique to create improved features for KWS. The experimental study is carried out for different test scenarios, windows and parameters, datasets, and neural networks commonly used in embedded KWS applications. Experiment results confirm the advantages of using the proposed improved features.