This article explores the possibility of using visual object detection techniques for word localization in speech data. Object detection has been thoroughly studied in the contemporary literature for visual data. Considering that an audio can be interpreted as a one-dimensional image, object localization techniques can be fundamentally useful for word localization. Based on this idea, we propose a lightweight solution for word detection and localization. We use bounding box regression for word localization, which allows our model to detect the occurrence, offset, and duration of keywords in a given audio stream. We experimented with LibriSpeech and trained a model to locate 1000 words. Compared to existing work (SpeechYolo), our method reduces the model size by 94% and improves the F1 score by 6.5%.
