5.9 Quantization of the adaptive and fixed codebook gains

26.1903GPPAdaptive Multi-Rate - Wideband (AMR-WB) speech codecSpeech codec speech processing functionsTranscoding functionsTS

The adaptive codebook gain (pitch gain) and the fixed (algebraic) codebook gain are vector quantized using a 6-bit codebook for modes 8.85 and 6.60 kbit/s and using a 7-bit codebook for all the other modes.

The fixed codebook gain quantization is performed using MA prediction with fixed coefficients. The 4th order MA prediction is performed on the innovation energy as follows. Let E(n) be the mean-removed innovation energy (in dB) at subframe n, and given by

( 49)

where N=64 is the subframe size, c(i) is the fixed codebook excitation, and dB is the mean of the innovation energy. The predicted energy is given by

( 50)

where [b₁ b₂ b₃ b₄]=[05.,0.4,0.3,0.2] are the MA prediction coefficients, and is the quantized energy prediction error at subframe k. The predicted energy is used to compute a predicted fixed-codebook gain g‘_c as in Equation (49) (by substituting E(n) by and g_c by g‘_c). This is done as follows. First, the mean innovation energy is found by

( 51)

and then the predicted gain g‘_c is found by

( 52)

A correction factor between the gain g_c and the estimated one g‘_c is given by

( 53)

Note that the prediction error is given by

( 54)

The pitch gain, g_p, and correction factor  are jointly vector quantized using a 6-bit codebook for modes 8.85 and 6.60 kbit/s, and 7-bit codebook for other modes. The gain codebook search is performed by minimizing the mean-square of the weighted error between original and reconstructed speech which is given

( 55)

where the x is the target vector, y is the filtered adaptive codebook vector, and z is the filtered fixed codebook vector. (Each gain vector in the codebook also has an element representing the quantized energy prediction error.) The quantized energy prediction error associated with the chosen gains is used to update . In the search, only the 64 codevectors that are closest to the unquantized pitch gain, g_p, are taken into account.