Multiple description dual linear decomposition transform

In this dissertation, a new speech coding algorithm called the multiple description dual linear decomposition transform (MDDLDT) is introduced. Based on the linear decomposition transform (LDT), the MDDLDT generates the optimum filter h₁ that best estimates the even-element sequence using the odd-element sequence and the optimum filter h ₂ that best estimates the odd-element sequence using the even-element sequence. In order to recover the lost data in a packet-switched network, the MDDLDT sends the optimum filter h₁ with the odd-element sequence as description 1 and the optimum filter h₂ with the even-element sequence as description 2. In the case where one of descriptions is lost, the lost data can be recovered using only information in the description that is received. Accordingly, three versions of the MDDLDT including the zero padding MDDLDT, the symmetric extension MDDLDT and the MDDLDT with sending outer segments are developed. Furthermore, the MDDLDT is modified to combat the consecutive packet loss case. The quality of recovered speech sequences using the MDDLDT is compared to existing recovery systems.