Processing and Recognition of Voice

Abstract

With rise of new technologies involving signal processing, the range of operations with signals and processing of those signals has become quite easy. Voice is considered to a unique feature of a person. So extraction of voice features and detecting and proces sing them in correct manner is always a matter of great concern. There’s been a lot of technique to detect the voice properly, but every method has some drawbacks due to some inherent property of voice. Voice can be considered to be a random signal with so me probabilities. So recognition of voice with good efficiency is not always an easy job to do. Here we have discussed the feature extraction of voice by MFCC model and checking those features by three different algorithms with efficiency comparison. The o bjective of this research was to make a comparative study between different signal matching techniques. Here samples of voice have been processed to get some certain features and the matching those features with one another to detect the similarity and get ting the information about the voice. Voice processing needs two steps. The first one is to train. Training includes feature extraction or processing of voice so that the processed data can be used for analytical purpose. Then the next step involves testin g, i.e. comparing the obtained data with one another to compare the similarity between two features obtained from processing of voice. Index Terms — DCT: Direct Cosine Transform DFT: Discrete Fourier Transform DWT: Dynamic time wrapping FT: Fourier Transfo rm HMM: Hidden Markov model MFCC: Mel frequency Cepstrum Coefficient VQ: Vector Quantization