Abstract:
The speech compression research aims to compress the speech signal while maintains the quality of the speech signal to similar as the original signal. Speech compression process, a noise might occur and effects to the quality of speech signal which can reduce the quality of the speech signal. This thesis provides the studying and exploring of Q and R parameters of Kalman filter for reducing the white Gaussian noise in speech compression.
Speech signal is passed through the Kalman filter and white Gaussian noise model at 10 Hz 50 Hz 100 Hz 500 Hz and 1000 Hz respectively. In the experiment, Q and R parameters of Kalman filter characteristics are adjusted, which effects to the unstable noise in the system, to reduce the noise. Moreover, it applied this technique to improve the quality of speech compression. The speech from Code Excite Linear Prediction speech compression, which constants of 10 male and 10 female speech signals, is tested with Kalman filter.
The results show the white Gaussian noise can be decreased when the Q and R parameters of Kalman filter is similar or equal. In the experiment, when Q and R are equal to 1, it can provide the maximum efficiency of white Gaussian noise reduction. In the condition of applying Kalman filter for improving the speech quality of CELP speech compression, this can reduce the low frequency noise and provides the better quality.