Muhammad Azam , pp. 93. ING/School of Engineering, 2011.
In packetized voice communication, speech packets are sometimes lost due to data transmission problems, e.g., signal fading, or interfering users and noise. For the recovery of missing speech packets, different methods are proposed. This thesis analyzes some recovery methods, and four variants of a waveform substitution method used during the objective analysis. This method is based on slow varying speech parameter estimates. These parameters include the short time energy (STE) and the zero crossing (ZC) measure. This technique is implemented in two different ways based on the slow varying parameters. These parameters are stored in the previous packet. If a speech packet is lost, it is recovered by the information stored in the previous packets. Both implementations differ only in the use of the zero crossing information. The short time energy estimation is the same in both implementations. A slight modification is made in these two implementations where the estimated speech parameters are stored in the previous and in future packets in order to recover two consecutive packets. This modification is applied only if the speech signal is already saved at the transmitter because it requires the future packets to store the information of previous packets, i.e., a non-causal solution. However, a causal solution is obtained if the signal is allowed to be delayed by one packet. The speech quality of the reconstructed speech signal is analyzed and compared between the four implementations. The implementation of these methods has been validated by subjectively observing the recovered speech packets, and by considering the improvement of the objective measures mean opinion score (MOS), mean square error (MSE) and signal-to-noise ratio (SNR). The recovery of samples within the packets is also discussed. The recovery of samples within a packet is done by the Fast Fourier Transform (FFT) block code method. The FFT block code method is implemented by an iterative algorithm. This method is validated by subjective observations and improvements in objective measures mean square error (MSE) and signal to noise ratio (SNR). The VAD is also used for the waveform substitution method and in the introduction of channel noise. After subjective observations and objective measures, it is concluded that modified method A provides better performance for the recovery of speech packets and the FFT block code method has been validated for recovering the samples within a packet.