Speak Search Shop - A Smart Voice Interaction Model for Modern E-Commerce

Modern eCommerce sites grapple with outdated search systems that use keywords to perform searches and provide a non-smooth product search experience, as well as restrict access to a wider range of user bases. In this study, Speak Search Shop is introduced, a voice-based product search model with React.JS as a frontend, Web Speech API as an orchestration microservice, and Elasticsearch as a contextual search microservice. The specification of requirements takes place in a form of a discussion; the user requests running shoes at a price below seven thousand rupees with an insole; these are then classified by intent and extracted into structured search terms. With 5,000 voice queries, experimental validation produced 94.2% intent classification accuracy, 91.7% entity extraction F1-score and average 687-milliseconds. Comparative analysis revealed enhanced first-result relevance by 37 percent, reduced zero-result queries by 42 percent, and reduced time to complete the task by 41 percent compared with text-based search. The 150-user test revealed a 33 percent increase in conversion rate and a significant boost in satisfaction, especially among the visually impaired users. The microservice containerized architecture embraces horizontal, Kube-based scaling, and can adjust to India English accent differences and code-switching habits. This architecture confirms the commercial viability of conversational interfaces as alternatives to traditional search, and shows that transformer-based language models can be integrated with production-scale infrastructure. Directions Future directions Multimodal search Multimodal search integrates voice and visual search, regional language support, and customized recommendation. The study brings reference designs of voice commerce systems to fit actual scale, latency, and precision requirements and improve accessibility in Internet retail settings.