[This article belongs to Volume - 54, Issue - 09]
Gongcheng Kexue Yu Jishu/Advanced Engineering Science
Journal ID : AES-14-11-2022-418

Title : Machine Learning Based Keyword Detection in a Multilingual Speech Dataset
Brajen Kumar Deka, Pranab Das,

Abstract : Speaking is the most common and practical form of communication. Automated voice recognition allows for natural communication between machines and people. It has developed into an interesting and challenging field. It enhances the computer’s capacity to react precisely to spoken words. A method for locating individual pronounced words in voice signals is called keyword spotting. Markov Chain Models without discrimination are commonly utilized in algorithm-finding keywords. This study presented a keyword identification method that used neural networks and iterative data to estimate keyword probabilities over time. This work aims to properly identify keywords in a recently constructed multilingual speech dataset. This dataset includes data in Hindi, English, and Assamese for 7-day, 10-digit, and 12-month periods. According to test findings, the suggested framework can correctly predict word samples in English, Assamese, and Hindi with 83.34 percent, 86.96 percent, and 81.36 percent accuracy during training. Applications for finding keywords in spoken text utilize isolated word recognition. This is useful for embedded and mobile devices.