Voice Assistant for Finding Items in Clothing Store

Abstract

The paper presents the use of machine learning in audio file processing, the use of neural networks to recognize voice commands, and the development of a voice assistant for finding products in a clothing store. 
 The purpose of the work is to develop a voice assistant for finding items in a clothing store.
 As a result, a dataset containing 30 categories and 3095 audio recordings was created, a neural network model was trained using the collected data, and an accuracy of 96.02% was achieved; WER: 0.0398; CER: 0.0087. The model was integrated with the search system into the voice assistant API, which allows you to record from a microphone, convert audio to text, break the text into keywords, and search the database using the obtained keywords. The speech recognition system has shown stable and high accuracy when recording from a microphone. This provides users with reliable and accurate recognition results even when using simple microphones. Search allows you to find results by keywords and names of items, and there is a function to recommend similar items if nothing was found at the user's request. The system's flexibility allows it to understand language contexts and does not depend on the order of words in a phrase, as the search is performed by keywords separately. The research analyzes the literature and compares existing approaches to voice command recognition. The methods of audio signal processing are described. The problem of searching for items in a clothing store was analyzed, and the current state of the market and demand for the technology were investigated. The practical value is that the developed voice assistant is specialized and optimized for searching for goods in a clothing store, allows solving the task of finding goods according to specified criteria, and simplifies the search task for different groups of people.