M.Sc. Tezi Görüntüleme | |||||||||||||||||||||
|
|
||||||||||||||||||||
| Summary: Speech is the most important means for the communication between the human beings since primitive ages. Mankind transmits their feeling, thoughts and desires to the others by using speech. People thought that there could be make a communication between a human and a computer like communication between human being and started to work to carry out this imagination. When you look the improvement on this topic, there are many considerable researches made in developed countries, especially in western countries. However, we could not say the same for the researches made for Turkish. In this study, our most important aim is aid to cover this gap. This study’s aim is making a phoneme based speech recognition and language identification system.Through the aim, a sound editor is firstly implemented to make sound records, investigate these records, extract the phonemes in these speech records and create our own phoneme database. To build the database, ten speech records (male and female) for each phoneme are taken from ten different persons. In addition to this, five Turkish and five English speech samples are also taken from the radio for the investigation the phonemes if they are exist or not in these records. In this study, we have used Word Detection, Fast Fourier Transform and Wavelet Transform for the spectral analysis of words and phonemes. Then, these spectral analysis results are examined by employing Cross Correlation technique and Dynamic Time Warping technique. Hence, the system could search the phonemes in the words and if the phoneme exists in the word, then it finds where it is and how many times the phone exists in this word. Finally, the obtained results are discussed in various respects. Keywords: Speech Analysis, Speech Recognition, Phoneme Recognition, Fast Fourier Transform, Haar Wavelet Transform, Word Detection, Dynamic Time Warping, Cross Correlation
| |||||||||||||||||||||