Transform-based audio codec and method with subband energy smoothing
A transform-based codec and method with energy smoothing for mitigating vector quantization errors (such as “birdies”) during the encoding process. Embodiments of the codec and method use an encoder to apply in combination an orthogonal transformation and a vector permutation to frequency transform coefficients. In some embodiments the transformation is performed first followed by the permutation and in other embodiments the order is reversed. The order used is reversed at the decoder. A smoothing parameter containing the level of energy smoothing to be applied is passed from the encoder to the decoder and used by both to compute a transform matrix and an inverse transform matrix. In some embodiments the transform matrix is a fraction Hadamard matrix that is invertible, energy preserving, controllable, and stable.
1. A method for processing an audio signal, comprising:
using an encoder to apply an orthogonal transformation and a vector permutation to frequency transform coefficients of the audio signal to generate processed frequency transform coefficients;
quantizing the processed frequency transform coefficients to obtain quantized processed frequency transform coefficients;
packing the quantized processed frequency transform coefficients into an encoded bitstream; and
using a decoder to process the encoded bitstream and reverse operations of the encoder and recover the audio signal.
2. The method of claim 1 , wherein the orthogonal transformation is any power of 2 greater than or equal to 4.
3. The method of claim 1 , further comprising applying the orthogonal transformation and the vector permutation to the frequency transform coefficients only once.
4. The method of claim 1 , further comprising applying the orthogonal transformation and the vector permutation to the frequency transform coefficients more than once.
5. The method of claim 1 , wherein the orthogonal transformation is variable and further comprising controlling the orthogonal transformation by a smoothing parameter that is a function at least one codec metric.
6. The method of claim 1 , wherein a size of a permutated sequence of the vector permutation is any value greater than or equal to a size of the orthogonal transformation.
7. The method of claim 1 , further comprising applying the vector permutation to a frequency transform coefficient vector containing the frequency transform coefficients to rearrange elements of the frequency transform coefficient vector according to a permutation rule, the frequency transform coefficient vector having a vector length.
8. The method of claim 7 , further comprising applying the orthogonal transformation, which is a transform matrix of size M, to the frequency transform coefficient vector.
9. The method of claim 8 , further comprising applying the transform matrix to M number of points at a time of the frequency transform coefficient vector, wherein the vector length is greater than or equal to M.
10. A method for encoding an audio signal to mitigate musical noise artifacts, comprising:
dividing the audio signal into a plurality of frequency subbands;
computing a transform matrix for each subband;
applying an orthogonal transformation to frequency transform coefficients in each subband using the transform matrix and applying interleaving to the frequency transform coefficients to obtain processed frequency transform coefficients, where the orthogonal transformation is any power of 2 greater than or equal to 4; and
vector quantizing and bit packing the processed frequency coefficients into an encoded bitstream representative of the audio signal such that musical artifacts are mitigated.
11. The method of claim 10 , further comprising:
computing a spectrum flatness for each subband; and
determining a smoothing parameter from the spectrum flatness that includes an energy smoothing level that represents an amount of energy smoothing to apply to each subband.
12. The method of claim 11 , further comprising transmitting the smoothing parameter in the encoded bitstream from an encoder to a decoder.
13. The method of claim 11 , further comprising using the smoothing parameter to compute the transform matrix.
14. The method of claim 13 , wherein computing a transform matrix for each subband further comprises:
computing a transform parameter as a function of the smoothing parameter, a length of a subband being processed, and a number of pulses of a vector quantizer; and
using the transform parameter to compute the transform matrix.
15. The method of claim 10 wherein the orthogonal transformation is applied first to the frequency transform coefficients and the interleaving is applied after the orthogonal transformation is applied.
16. The method of claim 10 wherein the interleaving is applied first to the frequency transform coefficients and the orthogonal transformation is applied after the interleaving is applied.