Speaker identification by means of wavelet and neural network aproach

The purpose of this study is to recognize the speaker identity by means ofwavelet analysis and neural network aproach. Firstly, sampling the voice signal generatedfrom 10 different person ( 6 males and 4 females ), we extracted noise from voice signals byusing Waveflow and Wavepad shareware programs. With the help of a Matlab Simulinkmodel we generated the useable data from this voice signals. Wavelet aproach has been usedfor getting these datas. These datas have been used as an input signal for Matlab basedNeural Network. This neural network classified the voice data for different speakers.In Section-1 ; The signal processing concept and it?s using area has been largelyexplained. Additionally, the advantages, disadvantages and the limitations of the speakerrecognition process have been criticized in the same section. Furthermore, some basicacknowledgements about the phases of speaker identification process have been presented inthis section. These phases are recording the voice signal, coding the signal, signal processingtechnics ( signal processing ) and signal modeling.In Section-2 ; For making easier the understanding of this study, new aproach tospeaker recognition, The process of HMM ( Hidden Markov Models ), theory and usage,Neural Networks used in modeling the voice signal, Time domain aproach to signalprocessing , LPC ( Linear Prediction Coding ) for signal modeling and coding, WaveletTheory and application areas have been underlined in this section.In Section-3 ; All the work done for this study; signal recording, signal processing,getting wavelet constants from the voice signal, neural network model preperation, workingwith the model and interpreating the output datas are briefly explained.In Section-4 ; The result and feature projects are explained.Key Words: Voice recognition, Neural Networks, Fast Fourier Transformation, Hidden Markov Models,Speaker Dependent Speaker Recognition, Wavelet