Vector quantization of LSP parameters using moving average interframe prediction

Abstract

This paper proposes a new method for vector quantization (VQ) of LSP parameters, using moving average (MA) interframe predictions. MA predictive VQ executes the prediction using the code vectors (codebook outputs) in the previous frames. Moreover, this method has the following features, compared to autoregressive (AR) predictive coding, which executes the prediction based on the previous quantized (decoded) values: (1) even if a bit error is produced in the transmission channel, its effect to the succeeding frames remains finite; and (2) the stored codes can be decoded from any time point. This paper discusses the quantization performance and the robustness against bit errors of the MA predictive VQ, as well as the training method for the codebook. Assuming an application to low‐bit‐rate speech coding, the configuration considered has a frame length of 40 ms, with each frame composed of four subframes. A method for further improvement of the efficiency is reported. As a result of evaluation using actual speech data, bit reductions of approximately 16 percent and 23 percent are achieved for the quantization for each 20 ms and 40 ms, respectively, compared to the conventional method which does not use interframe prediction.

Optimal Coding of Generalized-Gaussian-Distributed Frequency Spectra for Low-Delay Audio Coder With Powered All-Pole Spectrum Estimation

“…(4), (32), and (34) into eq. (3), as follows: Therefore, if we can prove , we can say that the two objective functions satisfy (37) and the equivalence of the problems will be proved. The rest of this section shows the proof for the statement .…”

“…(11). It should be noted that, since PAPSE is based on the all-pole model as is the conventional LP, the model parameters can be quantized using the conventional quantization algorithms for the LP coefficients such as in [37], i.e., the vector quantization of the coefficients in the form of Line Spectrum Pairs (LSP) [38].…”

“…The model parameters defined in eq. (11) were quantized by a total of 20 bits for each frame by using the method based on [37] with a different codebook created from a training data set for each condition, and the scale was logarithmically quantized into 8 bits. For the other model parameter , we assumed its ratio to the scale to be a constant that depends only on the bit rate and the shape parameter , since it is strongly related to the entropy of the generalized-Gaussian source.…”

A 5‐kHz‐bandwidth low‐bit‐rate speech coder: Bandwidth, noisiness, and opinion scores

“…After converting the 14th-order linear prediction coefficient obtained into LSP (Line Spectrum Pairs), a two-mode two-step vector quantizer [4,5] performs quantization using 4th-order moving average prediction. One bit is used in switching the moving average prediction mode.…”

4 kb/s multi-pulse based CELP speech coding using excitation switching

“…Two modes such as voiced and unvoiced modes are used in this study. Ten-th order filter LSP coefficients are vector quantized after MA (moving average) prediction [9], [lo]. In adaptive codebook, lag is differentially encoded [I I] in the second sub-frame in voiced frames, because lag is tempolarily correlated between sub-frames.…”