Machine Learning for Speech-to-Text: A Custom-CNN Model Approach

Abstract -Disabled persons can profit from its various potential applications and advantages, as it has a wide range of applications. Language problems prevent many individuals from communicating. With the intention of lowering this barrier, a machine-learning model was created in order to build systems that, in certain situations, can be a big assistance in enabling individuals to communicate information by using voice input to operate a computer. An easy-to-use and efficient voice recognition algorithm is described in this paper's research work. After converting the audio signal to the appropriate text, summarised text is produced. This paper presents research on a simple and efficient voice recognition technique. Summarised text is produced by converting the auditory signal to the appropriate text. The usage of machine learning in a variety of applications these days presents a challenge to model improvement. A custom-CNN model is developed to increase the precision of recognition. ReLU's quick computation makes it an excellent choice for CNN training. To confirm the effectiveness of the suggested approach, a great deal of experimentation is conducted. With an accuracy rate of 75.60%, the experimental results show how well the present CNN design performs.. Key Words: The Machine Learning, Speech-to-Text, Custom-CNN, Convolutional Neural Network, Audio Processing, Automatic Speech Recognition (ASR), Deep Learning, Feature Extraction.