Data processing: speech signal processing – linguistics – language – Speech signal processing – For storage or transmission
Reexamination Certificate
1999-10-06
2002-07-23
Korzuch, William (Department: 2641)
Data processing: speech signal processing, linguistics, language
Speech signal processing
For storage or transmission
C704S230000, C704S500000
Reexamination Certificate
active
06424939
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to a method for coding audio signals, and in particular to a method for coding an audio signal for the standard MPEG-2 AAC being just developed.
BACKGROUND OF THE INVENTION AND PRIOR ART
The standardization organization ISO/IEC JTC1/SC29/WG11, which is also known as Moving Pictures Expert Group (MPEG), was founded in 1988 in order to specify digital video and audio coding schemes for low data rates. The first specification phase was completed in November 1992 by the standard MPEG-1. The audio coding system according to MPEG-1, which is specified in ISO 11172-3, operates in a single-channel or two-channel stereo mode at sampling frequencies of 32 kHz, 44.1 kHz and 48 kHz. The standard MPEG-1 Layer II delivers broadcasting quality, as specified by the International Telecommunication Union, at a data rate of 128 kb/s per channel.
In its second development phase, the aims of MPEG consisted in defining multi-channel extension for MPEG-1-Audio, which was supposed to be backward-compatible with respect to existing MPEG-1 systems, and in defining also an audio coding standard at lower sampling frequencies (16 kHz, 22.5 kHz, 24 kHz) than with MPEG-1. The backward-compatible standard (MPEG-2 BC) as well as the standard with lower sampling frequencies (MPEG-2 LSF) were completed in November 1994. MPEG-2 BC delivers a good audio quality at data rates from 640 to 896 kb/s for 5 channels with full bandwidth. Since 1994, further endeavors of the MPEG-2 audio standardization committee consist in defining a multi-channel standard of higher quality than that obtainable when backward compatibility to MPEG-1 is demanded. This not backward-compatible audio standard according to MPEG-2 is designated MPEG-2 NBC. The aim of this development consists in obtaining broadcasting qualities according to the ITU-R requirements at data rates of 384 kb/s or lower rates for 5-channel audio signals in which each channel has the full bandwidth. The audio coding standard MPEG-2 NBC was completed in April 1997. The scheme MPEG-2 NBC will constitute the core for the already planned audio standard MPEG-4 that will have higher data rates (above 40 kb/s per data channel). The NBC or not backward-compatible standard combines the coding efficiency of a high-resolution filter bank, prediction techniques and redundancy-reducing Huffman coding in order to obtain audio coding of broadcasting quality at very low data rates. The standard MPEG-2 NBC is also referred to as MPEG-2 NBC AAC (AAC=Advanced Audio Coding). A detailed representation of the technical contents of MPEG-2 AAC can be found in M. Bosi, K. Brandenburg, S. Quackenbush, L. Fielder, K. Akagiri, H. Fuchs, M. Dietz, J. Herre, G. Davidson, Yoshiaki Oikawa: “ISO/IEC MPEG-2 Advanced Audio Coding”, 101st AES Convention, Los Angeles 1996, Preprint 4382.
Efficient audio coding methods remove both redundancies and irrelevancies from audio signals. Correlations between audio sampling values and statistics of sample value representation are exploited for removing redundancies. Frequency range and time range masking properties of the human hearing system are exploited in order to remove non-perceptible signal components (irrelevancies). The frequency content of the audio signal is subdivided into partial bands by means of a filter bank. The data rate reduction is achieved by quantizing the spectrum of the time signal in accordance with psycho-acoustic models, and this reduction may comprise a loss-free coding method.
Generally speaking, a time-continuous audio signal is sampled in order to obtain a time-discrete audio signal. The time-discrete audio signal is subjected to windowing by means of a window function in order to obtain successive blocks or frames of a specific number of, e.g. 1024, windowed time-discrete sampling values. Each block of windowed time-discrete sampling values of the audio signal is successively transformed to the frequency range, for which a modified discrete cosine transformation (MDCT) can be employed for example. The thus obtained spectral values are not yet quantized and thus need to be quantized. The main object in doing so consists in quantizing the spectral data in such a manner that the quantization noise is masked or covered by the quantized signals themselves. For doing so, a psycho-acoustic model described in the MPEG AAC standard is employed which, in consideration of the specific properties of the human ear, calculates masking thresholds depending on the audio signal present. The spectral values then are quantized in such a manner that the quantization noise introduced is hidden and thus becomes inaudible. Quantization thus does not result in audible noise.
The NBC standard makes use of a so-called non-uniform quantizer. In addition thereto, a method is used for shaping the quantization noise. The NBC method uses, just as preceding standards, the individual amplification of groups of spectral coefficients that are known as scale factor bands. In order to operate in as efficient manner as possible, it is desirable to be able to shape the quantization noise in units that are as far as possible matched to the frequency groups of the human hearing system. It is thus possible to group spectral values that very closely reflect the bandwidth of the frequency groups. Individual scale factor bands can be amplified by means of scale factors in steps of 1.5 dB. Noise shaping is achieved as amplified coefficients have higher amplitudes. Therefore, they will in general display a higher signal-to-noise ratio after quantization. On the other hand, higher amplitudes require more bits for coding, i.e. the bit distribution between the scale factor bands is changed implicitly. The amplification by the scale factors, of course, has to be corrected in the decoder. For this reason, the amplification information stored in the scale factors in units of 1.5 dB steps must be transmitted to the decoder as side information.
After quantization of the spectral values present in scale factor bands and optionally amplified by scale factors, the spectral values are to be coded. The input signal to a noise-free coding module thus is the set of, for example, 1024 quantized spectral coefficients. By means of the noise-free coding module, the sets of 1024 quantized spectral coefficients are divided into sections in such a manner that a single Huffman coding table (“codebook”) is employed for coding each section. For reasons of coding efficiency, section limits can be present only on scale factor band limits, such that for each section of the spectrum both the length of the section in scale factor bands and the Huffman coding table number used for the section must be transmitted as side information.
The formation of the sections is dynamic and varies typically from block to block, such that the number of bits necessary for representing the full set of quantized spectral coefficients is reduced to a minimum. Huffman coding is used for representing n-tuples of quantized coefficients, with the Huffman Code being derived from one of 12 coding tables. The maximum absolute value of the quantized coefficients that can be represented by each Huffman coding table as well as the number of coefficients in each n-tuple for each coding table are defined from the very beginning.
The reason for forming the sections thus consists in grouping regions having the same signal statistics, in order to obtain an as high as possible coding gain for a section by means of one single Huffman coding table, with the coding gain being defined in general by the quotient of bits prior to coding and bits after coding. By means of a coding table number (“codebook number”) defined in the bit flow syntax used for the NBC method, reference is made to one of the 12 Huffman coding tables which permits the highest coding gain for a specific section. The expression “coding table number” in the present application thus is to refer to the location in the bit flow syntax reserved for the coding table number. For coding 11 different coding table numbers in b
Brandenburg Karlheinz
Dietz Martin
Ehret Andreas
Gbur Uwe
Gerhauser Heinz
Beyer Weaver & Thomas LLP
Fraunhofer-Gesellschaft zur Forderung Der Angewandten Forschung
Korzuch William
McFadden Susan
LandOfFree
Method for coding an audio signal does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for coding an audio signal, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for coding an audio signal will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2846366