AGILE-KWS · A Giant Leap for Keyword Spotting

Speech technologies like keyword spotting (KWS), which is concerned with the detection of keywords from an audio stream comprising speech, are nowadays in vogue thanks to their integration in many electronic devices in order to comfortably interact with them by means of the voice. Despite all the research progress made over the last decades, the performance of speech technologies is still severely degraded when acoustic distortions like background noise or reverberation are present. This can make these technologies completely unusable for the real-world applications. As a result, this project proposal aims at taking KWS technology forward by researching and developing novel state-of-the-art acoustic distortion-robust methods based on concepts successfully applied to related areas like automatic speech recognition to get a leap of KWS performance in real-world (i.e., noisy) acoustic conditions. This will be achieved while working towards two different and exciting timely goals with high social impact giving an interdisciplinary dimension to this action: 1) the preservation of the NASA’s Apollo mission speech recordings for future generations (during the outgoing phase), and 2) the improvement of hearing-impaired people’s life quality by facilitating voice-based interaction with hearing assistive devices like hearing aids (during the incoming phase).