Inventor(s)

Abstract

The codec works on a group of 4 samples per channel (83.3 µs at 48 kHz) and emits a fixed number of bits for every group. There is no frame, no bit reservoir, no lookahead, and no adaptation that spans groups other than backward-adaptive state the decoder tracks from the bitstream alone. A two-stage QMF tree splits each channel into four 6 kHz bands; the default tree uses a 16-tap filter at the 12 kHz split and 8-tap filters at the two lower splits (26 samples, 0.54 ms of filter delay). Each band is coded by its own ADPCM coder with a fixed bit depth, a µ-law companded quantiser, a pole-zero predictor adapted by leaky sign-sign LMS, and a leaky log-domain step-size adaptation, so any mismatch between encoder and decoder decays. A stereo codec chooses between mid/side, left/right and right/left coding backward-adaptively from the decoded signal, at no bit cost; its choice as each block starts (2 bits) is carried by the link in the block descriptor, which is inside the block’s error-correcting codewords and its keyed tag. Every block also carries, for each stereo mix, a 5-bit step-size field holding one band’s quantised log step, the band chosen by the block number. The field sits right after the block descriptor, before the first audio group, and is part of the mix’s error-correcting codeword and of the keyed tag’s input.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS