Spoken Metro Station Name Identification: A Deep Learning-Based Approach

Tourism is an up-and-coming industry with a significant source of income for the states and central governments all over the world. It also drives the livelihood of multitudinous locals in tourist spots. One of the problems often faced by tourists is navigation through cities’ tourist spots. Signboards, banners, and informative texts are often written in local languages and English (at times). This poses several difficulties for travelers who neither know local languages nor English. They encounter a daunting challenge when trying to navigate within the local area, and frequently become victims of dishonest individuals who exploit their lack of knowledge. This ultimately paints a dark picture of a place in front of the world. Voice-based systems can be beneficial in this context. These systems can enable visitors to query about different places, get directions, know about attractions, and other to-do things in a city. They can get accurate answers by just “asking” about a place from the system, thus avoiding the need for reading/writing ability of the dominant languages of that place. This can furthermore help impaired people in their daily travel. This paper proposes a deep learning-based approach with deep learning to address some of the above-mentioned issues. At the outset, the system is trained to recognize the metro station names in Kolkata (Capital city of West Bengal, India) from speech. This functionality can not only help tourists to navigate in the city but also aid in speeding up the ticketing system within metro stations by introducing voice-based input to the automated ticket vending machines. To evaluate the proposed system, several experiments were performed on a dataset of 24 metro station names in Kolkata, and the best accuracy of over 95% successful recognition was obtained in non-studio conditions.