Compression of audio scale-factors by two-dimensional transformation
View Patent ↗Digital audio samples are represented as a product of scale factors codes and corresponding quantity codes, sometimes referred to as exponent/mantissa format. To compress audio data, scale factors are organized by sample time and frequency either by filtering or frequency transformation, into a two-dimensional frame. The frame may be decomposed into “tiles” by partition. One or more such scale factor tiles are compressed by transformation by a two-dimensional, orthogonal transformation such as a two dimensional discrete cosine transform. Optional further encoding is applied to reduce redundancy. A decoding method and an encoded machine readable medium complement the method of encoding.
1. A method of compressing a digitized audio signal representing a sound, said signal having an audio bandwidth, in an audio compression system representing sound samples or spectral values using a scale-factor plus data format, wherein a sample is represented as a product of a scale factor and an associated quantity—the method comprising the steps of:
receiving a digital signal representing a sound;
organizing samples into at least one audio frame, said frame comprising a plurality of temporally sequential samples representing a time interval;
for each frame, processing said plurality of temporally sequential samples into a plurality of subband signals, each subband signal representative of a respective subband frequency range and comprising a time sequence of audio samples within said subband frequency range;
converting said subband signals into a format expressing each filtered audio sample as a product of a) a scale factor, represented in a scale factor field and b) a quantity, represented in a quantity field;
organizing in two dimensions the scale factor fields of said subband signals at least one tile corresponding to each frame said tile comprising a matrix of scale factors organized by time as a first dimension, and a subband frequency range as a second dimension;
processing said at least one tile with a two dimensional orthogonal transform to produce for each said tile a respective scale factor coefficient matrix (SCM);
compressing each said SCM to produce a compressed coefficient matrix representing the scale factors in a tile in a compressed format;
packing said compressed coefficient matrix in a data format for transmission.
2. The method of claim 1 , wherein said orthogonal transform comprises a two dimensional, discrete cosine transform.
3. The method of claim 1 , wherein said at least one tile comprises a plurality of tiles, said plurality of tiles derived by partition of a two-dimensional matrix representing a complete audio frame;
each said tile representing a sub-interval of time and a fraction of the frequency range of said complete audio frame.
4. The method of claim 3 , further comprising the step of:
after said step of processing each of said tiles, requantizing said at least one tile in accordance with a requantization matrix.
5. The method of claim 1 , wherein said step of compressing comprises:
for at least one SCM, rearranging coefficients into a string of coefficients.
6. The method of claim 5 , wherein said step of compressing further comprises:
using an entropy reducing code to compress said string of coefficients.
7. The method of claim 6 , wherein said entropy reducing code comprises a Huffman code.
8. The method of claim 1 , wherein said compressing step includes using differential coding across related tiles in a common frame.
9. The method of claim 1 , wherein said step of organizing the scale factor fields comprises modifying said at least one tile by a prediction model that models a matrix by a calculated trend across at least one of a) rows, and b) columns, to obtain a modified matrix of scale factors.
10. The method of claim 9 , wherein said prediction model comprises a linear prediction model, and wherein said calculated trend is a linear trend.
11. The method of claim 9 , wherein said prediction model comprises a polynomial model, and said calculated trend comprises a polynomial function.
12. The method of claim 1 , wherein said step of processing said plurality of temporally sequential samples into a plurality of subband signals comprises:
filtering said temporally sequential samples with a bank of digital bandpass filters, then decimating to generate a plurality of critically sampled subband signals.
13. The method of claim 1 , wherein said step of processing said plurality of temporally sequential samples into a plurality of subband signals comprises:
transforming sequential sets of said samples into a frequency domain representation by a frequency transform, to produce for each said set a series of subband signals corresponding to a set of frequency bins.
14. The method of claim 1 , further comprising the step of transmitting said compressed coefficient matrix through a transmission medium.
15. The method of claim 14 , wherein said transmission medium includes a data network.
16. The method of claim 1 , further comprising the step of recording said compressed coefficient matrix on a machine readable medium.
17. A method of decoding an encoded electronic data signal representing an audio signal, useful in decoding a signal wherein samples are encoded by a system representing sound samples or spectral values using a scale-factor plus quantity format, wherein a sample is represented as a product of a scale factor and an associated quantity Q—the decoding method comprising the steps of:
decoding an input signal into data packets;
unpacking a received data packet to separate encoded scale factor data and encoded quantity data;
decompressing said encoded scale factor data to generate at least one coefficient matrix; and
transforming said at least one matrix by a two dimensional inverse orthogonal transform, said inverse orthogonal transform the inverse of an orthogonal transform used to encode said coefficient matrices, to obtain at least one corresponding scale factor matrix.
18. The method of claim 17 , wherein said inverse orthogonal transform comprises an inverse, two-dimensional discrete cosine transform.
19. The method of claim 17 , wherein:
said at least one coefficient matrix comprises a plurality of coefficient matrices;
said step of transforming said at least one coefficient matrix comprises transforming each of a plurality of coefficient matrices to obtain a plurality of corresponding scale factor submatrices;
and further comprising the step of assembling said scale factor submatrices into a larger frame matrix, by concatenating said scale factor submatrices in a predetermined pattern of tiles corresponding to a tiling pattern used in a known encoder.
20. The method of claim 17 , wherein said step of decompressing said encoded scale factor data comprises decoding an entropy reducing code.
21. The method of claim 20 , wherein said entropy reducing code comprises a Huffman code.
22. The method of claim 21 , wherein said step of decompressing said encoded scale factor data further comprises decoding differences between adjacent submatrices in a common frame matrix, and summing said differences to reconstruct submatrices.
23. The method of claim 21 , further comprising the step of:
re-quantizing said scale factor matrix to obtain a decompressed, requantized scale factor matrix by converting said decompressed scale factors from a non-linear quantization to a linear scale factor, thereby calculating a scale factor matrix for an audio frame.
24. The method of claim 17 , further comprising the steps of:
multiplying elements of said scale factor matrix by corresponding data quantities (Q) to reconstruct a matrix of audio samples.
25. The method of claim 17 , wherein said input signal is received from a data network.
26. The method of claim 17 , wherein said signal is read from a machine readable storage medium.
27. A method of decoding an encoded electronic data signal representing an audio signal, useful in decoding a signal wherein samples are encoded by a system representing sound samples or spectral values using a scale-factor plus quantity format, wherein a sample is represented as a product of a scale factor and an associated quantity Q—the decoding method comprising the steps of:
unpacking a received data packet to separate encoded scale factor data and encoded quantity data;
decompressing said encoded scale factor data to generate at least one coefficient matrix;
transforming said at least one matrix by a two dimensional inverse orthogonal transform, said inverse orthogonal transform the inverse of an orthogonal transform used to encode said coefficient matrices, to obtain at least one corresponding scale factor matrix.
multiplying elements of said scale factor matrix by corresponding data quantities (Q) to reconstruct a matrix of audio samples; and
processing said matrix of audio samples to construct a stream of sequential, digitized audio samples.
28. The method of claim 27 , wherein said step of processing said matrix of audio samples comprises:
for each row of said matrix of audio samples, processing the row with a synthesizing filter, said synthesizing filter up-sampling the row to a frame length in samples, in a manner complementary to a decimation performed in a known encoder, thereby obtaining a plurality of reconstructed subband signals
mixing said reconstructed subband signal to produce a replica of a full band audio signal representing a sound.
29. The method of claim 28 , further comprising the step of outputting said audio samples to another device for purposes of reproducing a sound.
30. The method of claim 28 , further comprising the step of causing a sound to be reproduced, based upon said stream of sequential digitized audio samples, said sound an approximate replica of a sound encoded by a method compatible with the method of decoding.