Efficient speaker identification from speech transmitted over bluetooth based system

This paper presents a study for speaker recognition of the speech signals transmitted through Bluetooth channel as degraded speech signals, while the training phase is made with clean speech signals. This is based on the Mel-Frequency Cepstral Coefficients (MFCCs) for feature extraction from the speech signals. Different approaches for feature extractions are tested in the paper; feature extraction from the signals, feature extraction from the Discrete Cosine Transform (DCT) of signals, feature extraction from the signals and the DCT, feature extraction from the Discrete Sine Transform (DST) of signals, feature extraction from the signals and the DST, feature extraction from the Discrete Wavelet Transform (DWT) of signals, and finally feature extraction from the signals and the DWT. A Neural Network (NN) classifier is used in the simulation experiments. Simulation results show that feature extraction from the DCT of signals achieves the highest recognition rates.