Speech to Text Synthesis from Video Automated Subtitling using Levinson Durbin Method of Linear Predictive Coding

摘要

The objective of speech processing is to make sure that the information to be transferred is clear and accurate, although currently many algorithms have been developed; challenge still exists in extracting the features and reconstructing it to reproduce. Feature extraction is the first thing for speech processing so that a digital system will know what is the word exactly spoken by people. In this work, a Graphical User Interface (GUI) has been developed using matlab to extract the audio speech from the test video. Noise is removed by designing a Butterworth filter and frame by frame analysis is done to separate words in the speech using a threshold. Feature extraction is done using Levinson-Durbin method of Linear Predictive Coding and given as input to the neural network for training and recognition. Results are quite satisfied with overall recognition rate of 76%.