System and method for long term prediction in audio codecs
A frequency domain long-term prediction system and method for estimating and applying an optimum long term predictor. Embodiments of the system and method include determining parameters of a single-tap predictor using a frequency-domain analysis having an optimality criteria based on spectral flatness measure. Embodiments of the system and method also include determining parameters of the long-term predictor by accounting for the performance of the vector quantizer in quantizing the various subbands. In some embodiments other encoder metrics (such as signal tonality) are used as well. Other embodiments of the system and method include determining the optimal parameters of the long-term predictor by accounting for some of the decoder operation. Other embodiments of the system and method include extending a 1-tap predictor to a k-th order predictor by convolving the 1-tap predictor with a pre-set filter and selecting from a table of such pre-set filters based on a minimum energy criteria.
1. An audio coding system for encoding an audio signal, comprising:
a frequency transformation unit that represents the windowed time signal in a frequency domain to obtain a frequency transformation of the audio signal;
an optimal long-term predictor estimation unit that estimates long-term predictor coefficients based on an analysis of the frequency transformation and a criteria of optimality in the frequency domain;
a long-term predictor that filters the audio signal in the time domain, wherein the long-term predictor is an adaptive filter with coefficients that are the long-term predictor coefficients estimated from the analysis performed by the optimal long-term predictor estimation unit in the frequency domain;
a quantization unit that quantizes frequency transform coefficients of a windowed frame to be encoded to generate quantized frequency transform coefficients; and
an encoded signal containing the quantized frequency transform coefficients, and where the encoded signal is a representation of the audio signal.
2. The audio coding system of claim 1 , wherein the optimal long-term predictor estimation unit further comprises estimating the optimal long-term linear predictor based on an analysis of a quantization error from the quantization unit.
3. The audio coding system of claim 1 , further comprising:
a filter shapes table of pre-determined filter shapes used to extend a 1-tap long-term linear predictor into a k-th order long-term linear predictor; and
an estimation selection unit that selects the optimal filter shape from the filter shapes table.
4. The audio coding system of claim 3 , further comprising the optimal filter shape that is selected by minimizing an energy of an output of the k-th order long-term linear predictor.
5. A method for encoding an audio signal, comprising:
generating a frequency transformation for the audio signal, the frequency transform representing a windowed time signal in a frequency domain;
estimating long-term predictor coefficients based on an analysis of the frequency transformation and a criteria of optimality in the frequency domain;
filtering the audio signal in the time domain using a long-term linear predictor, wherein the long-term linear predictor is an adaptive filter with coefficients that are the long-term predictor coefficients that were estimated from the analysis in the frequency domain;
quantizing frequency transform coefficients of a windowed frame to be encoded to generate quantized frequency transform coefficients; and
constructing an encoded signal containing the quantized frequency transform coefficients, wherein the encoded signal is a representation of the audio signal.
6. The method of claim 5 , further comprising determining adaptive filter coefficients for the long-term linear predictor based on a frequency analysis of a windowed time signal of the audio signal.
7. The method of claim 5 , further comprising estimating the optimal long-term linear predictor based on both the analysis of the frequency transformation and a quantization error from quantization of the frequency transformation coefficients.
8. The method of claim 5 , further comprising:
extending a 1-tap long-term linear predictor into a k-th order long-term linear using a predictor filter shapes table containing pre-determined filter shapes; and
selecting an optimal filter shape from the predictor filter shapes table for use in the optimal long-term linear predictor.
9. The method of claim 8 , wherein selecting the optimal filter shape further comprises selecting a filter shape from the predictor filter shapes table that minimizes an energy of an output of the k-th order long-term linear predictor.
10. The method of claim 5 , wherein the long-term linear predictor is a 1-tap long-term linear predictor and further comprising estimating lag and gain parameters for the 1-tap long-term linear predictor.
11. The method of claim 10 , further comprising:
determining dominant peaks in a frequency magnitude spectrum corresponding to the dominant harmonic components in the windowed time signal and computing a fractional frequency for each of the dominant peaks;
constructing a set of candidate filters in the frequency domain based on a subset of the dominant peaks and applying this set of candidate filters to the frequency magnitude spectrum to generate a resultant transform spectrum; and
computing the criteria of optimality.
12. The method of claim 11 , further comprising wherein the frequency-based criteria of optimality is the spectral flatness measure of the resulting spectrum after applying the candidate filter:
selecting the optimal filter shape that maximizes the criteria of optimality;
converting the lag and gain parameters determined in a frequency analysis into a time-domain equivalent; and
applying, in the time domain to the audio signal, the optimal long-term linear predictor containing the lag and gain parameters, wherein the optimal filter shape contains the lag and gain parameters.
13. The method of claim 11 , further comprising quantizing the resultant transform spectrum using a scalar or a vector quantizer;
generating a measure of the quantization error for a selected bit rate; and
estimating the optimal long-term linear predictor based on a combination of a measure of the quantization error and spectral flatness measure.
14. The method of claim 13 , further comprising imposing an upper limit on a gain of the optimal long-term linear predictor using the quantization error and a frame tonality measure.
15. The method of claim 14 , further comprising estimating the optimal long-term linear predictor based on minimizing reconstruction signal error at the decoder.
16. A method for encoding an audio signal, comprising:
filtering the audio signal using a long-term linear predictor, wherein the long-term linear predictor is an adaptive filter;
generating a frequency transformation for the audio signal, the frequency transform representing a windowed time signal in a frequency domain;
estimating an optimal long-term linear predictor based on an analysis of the frequency transformation and a criteria of optimality in the frequency domain;
extending a 1-tap long-term linear predictor into a k-th order long-term linear using a predictor filter shapes table containing pre-determined filter shapes;
selecting an optimal filter shape from the predictor filter shapes table that minimizes an energy of an output of the k-th order long-term linear predictor for use in the optimal long-term linear predictor;
quantizing frequency transform coefficients of a windowed frame to be encoded to generate quantized frequency transform coefficients; and
constructing an encoded signal containing the quantized frequency transform coefficients, wherein the encoded signal is a representation of the audio signal.
17. The method of claim 16 , further comprising determining adaptive filter coefficients for the long-term linear predictor based on a frequency analysis of a windowed time signal of the audio signal.
18. The method of claim 16 , further comprising estimating the optimal long-term linear predictor based on both the analysis of the frequency transformation and a quantization error from quantization of the frequency transformation coefficients.