الفهرس | Only 14 pages are availabe for public view |
Abstract Reverberation effect is scientifically defined as the presence of sound in closed rooms after removal of the sound source. It occurs due to multiple reflections from the room walls, ceiling and ground. Both the shape and size of the room affect the reverberation that occurs in the room. Reverberation generally spreads the energy of sound. This, in turn, changes the speech signal characteristics such as pith frequency. It is known that the pitch frequency is the fundamental frequency in the signal, and it is necessary to be determined accurately for several applications. Different methods have been introduced in the literature for pitch frequency estimation. These methods include Normalized Correlation Function (NCF), Cepstrum Pitch Detection (CEP), Summation of Residual Harmonics (SRH) and Pitch Estimation Filter (PEF). This thesis is concerned with the investigation of reverberation effect on pitch frequency estimation. The studied methods for pitch frequency estimation are compared in the presence of reverberation. It is found that the PEF method is preferred for pitch frequency estimation in the presence of reverberation. In addition, speaker identification is investigated in this thesis in the presence of reverberation. Deep neural networks are investigated for this task as they are efficient tools for feature extractions and classification. Speech signals are first transformed to spectrograms, and then features are extracted from these spectrograms. Simulation results proved good results for speaker identification in the presence of the reverberation effect. Both text-dependent and text-independent recognition systems have been presented and studied in the presence of degradation phenomena such as noise and reverberation. The experimental results reveal that the recognition rates obtained for text-dependent speaker recognition are higher than those of text-independent speaker recognition. |