LEVERAGING SPEECH FOR DYNAMIC IMAGE CAPTIONING: A MOBILENETV3 AND LSTM APPROACH

This paper presents a novel way of generating captions for images using an automatic image captioning system; the proposed model combines MobileNetV3 and LSTM to create captions that are accurate and relevant to the image being depicted. MobileNetV3 acts as a feature extractor as it extracts vital components of pictures at a reasonable computational cost. These features are then taken to the LSTM network and from it, descriptive captions from visual context are made. As a measure that improves user access, the generated captions are further translated to sound using Google Text-to-Speech (GTTS), which is especially important for the visually impaired and other hand-free users. Cross-sectional experimental assessments of the performances of the proposed model on the Flickr8k dataset further validate the impressive usefulness of the proposed model for generating faithful and comprehensive descriptions for media items that are potentially useful in assistive technologies, media organizing, and interactive systems.