5.2.4 Adaptive interpolation of LSFs
26.4473GPPCodec for Enhanced Voice Services (EVS)Error concealment of lost packetsRelease 17TS
The sub-frame LSFs are interpolated based on, and using fixed interpolation factors given by
(23)
where and correspond to the mid and end LSFs of frame n respectively. Note that and such that, and those are fixed values known to both encoder and decoder. If the frame is lost its end LSFs are estimated by the decoder. However the dependence on the estimated end LSFs of the th frame may adversely affect the speech quality if the estimated end LSFs are not well represent the actual one. This issue is addressed by selecting the interpolation factors and appropriately by giving more weight to the end LSF of frame which is not lost.
To adaptively select the LSF interpolation factors, we estimate the gain of the synthesis filter resulting from the LSF vectors and by computing the energy of the impulse response of the corresponding synthesis filters. Let the impulse responses of the synthesis filters corresponding to and are given by and . The truncated energy of the impulse responses are given by and where
(24)
Note that N is the length of the truncated response. Typically 128 samples are used to compute the truncated impulse response. The interpolation factorsand are picked based on the energy ratio, coder type, FEC classification and the use of safety net quantization for LSF quantization.
These interpolation factors are picked from four different sets of interpolation factors. For example a largest difference in should pick the interpolation factors that give the very little or zero weight to the previous end LSF in the interpolation.
Figure 1: Adaptive interpolation of LSFs