Wav2KWS: Transfer Learning From Speech Representations for Keyword Spotting

With the expanding development of on-device artificial intelligence, voice-enabled devices such as smart speakers, wearables, and other or edge processing systems have been proposed. However, building obtaining large training datasets that are essential for robust keyword spotting (KWS) remains cumbersome. To address this problem, we propose a deep neural network can rapidly establish high-performance KWS system from arbitrary instruction sets. We use an encoder pretrained with large-scale speech corpus backbone then design effective transfer KWS. demonstrate feasibility proposed network, various experiments were conducted on Google Speech Command Datasets V1 V2. In addition, to verify applicability different languages, using three Korean command datasets. The outperforms state-of-the-art networks in both experiments. Furthermore, understand real human voice even when trained synthetic text-to-speech data.