An MDCT Domain Frame-Loss Concealment Technique for MPEG Advanced Audio Coding

Abstract

We propose a frame loss concealment technique for decoders compatible with MPEG advanced audio coding (AAC). The spectral information of the lost frame is estimated in the modified discrete cosine transform (MDCT) domain via efficient techniques that are tailored to individual source signal components: In noise-like spectral bins the MDCT coefficients are obtained by shaped-noise insertion, while coefficients in tone-dominant bins are estimated by frame interpolation followed by a refinement procedure so as to optimize the fit of the concealed frames with neighboring frames. Experimental results demonstrate that the proposed technique offers performance superior to techniques adopted in commercial AAC decoders.

Bidirectional cascaded long term prediction for frame loss concealment in polyphonic audio signals

“…SSNR results for each FLC technique, evaluated for all the files is given in Table 1. The table clearly shows that the lost frame reconstructed via the proposed FLC technique is closest to the original frame, with an average segmental SNR improvement of on the average 6.82 dB over previously known best technique described in [5]. Note that the poor SSNR results of the competitive methods is mainly because their objective is not to absolutely match the waveform of the lost frame and have sections of MDCT coefficients adjusted with random signs.…”

“…• Reference decoder with MDCT domain energy interpolation FLC module as proposed in [5] (further referred as MDCT-FLC)…”

“…Note that to have M f = K/2 future samples, 2 future frames have to be available to the FLC module. This requirement is same as the energy interpolation method proposed in [5], while more than the sub-band domain prediction based method proposed in [3], as it requires only one future frame to adjust energy within each band of the lost frame. The sub-band domain prediction based technique has been observed to result in poor concealment quality when compared to the method proposed in [5] and we hypothesize this to be due to the fact that the prediction does not account for smooth transition into future samples.…”

“…This requirement is same as the energy interpolation method proposed in [5], while more than the sub-band domain prediction based method proposed in [3], as it requires only one future frame to adjust energy within each band of the lost frame. The sub-band domain prediction based technique has been observed to result in poor concealment quality when compared to the method proposed in [5] and we hypothesize this to be due to the fact that the prediction does not account for smooth transition into future samples. Thus we emphasize on having future samples available and for this we need at least 2 future frames, as with only one future frame no future samples can be reconstructed due to the overlapped frames and the aliasing introduced during MDCT.…”

“…This approach does not utilize samples from future frames and is effectively an extrapolation technique with the shortcoming that it disregards smooth transition into future frames. An alternative approach to perform FLC in the modified discrete cosine transform (MDCT) domain, which accounts for future frames, was developed in our group [5]. This technique isolated tonal components in MDCT domain and interpolated the relevant missing MDCT coefficients of the lost frame using available past and future frames.…”