Data processing: speech signal processing – linguistics – language – Speech signal processing – For storage or transmission
Patent
1996-06-24
1999-08-31
Hudspeth, David R.
Data processing: speech signal processing, linguistics, language
Speech signal processing
For storage or transmission
341126, H04B 1404
Patent
active
059466529
DESCRIPTION:
BRIEF SUMMARY
FIELD OF THE INVENTION
The invention relates to signal compression, in which an information signal is represented using a reduced number of bits, and to signal expansion in which the number of bits representing the information signal is restored.
BACKGROUND OF THE INVENTION
When a digital information signal is compressed for storage and/or transmission at a reduced bit rate, the samples of the information signal are often requantized using a reduced word length. Reducing the word length by which each of the samples is represented reduces the number of bits required to represent the information at the expense of an increase in quantizing errors. This requantizing is usually done linearly by uniformly distributing quantizing values corresponding to the available quantizing levels across the range of values of the information signal and requantizing each sample of the information signal to the closest quantizing value. The same technique may be used to requantize transform coefficients representing the information signal.
In some applications, such as the so-called ISO/MPEG layer three audio coding standard, the quantizing values corresponding to the available quantizing levels are distributed non-uniformly. Distributing the quantizing values non-uniformly allows certain values of the information signal to be quantized more accurately than other values. This provides better results than linear quantizing when, for example, the values that are quantized more accurately occur more frequently than the values that are quantized less accurately, or errors in the values that are quantized more accurately are more noticeable. Both of these possibilities may occur in low bit rate audio compression.
Many audio compression systems orthogonally transform the audio input signal into the frequency domain before quantizing. When this is done, small-amplitude spectral coefficients occur more frequently than large-amplitude spectral coefficients for most practical input signals. Additionally, because of the psychoacoustic phenomenon of simultaneous masking that occurs between the members of a group of contiguous spectral coefficients, noise due to quantizing errors is less likely to be heard in the presence of large-amplitude coefficients than in the presence of small-amplitude coefficients. For this reason, a non-linear quantizing scheme that concentrates more quantizing levels at small amplitudes can lead to a significant improvement in sound quality.
Non-linear quantizing is normally implemented by a applying linear quantizing to a non-linearized version of the data subject to quantizing. The data subject to quantizing are normally a block or set of data, such as a set of spectral coefficients, that are to be quantized using the same number of quantizing levels, or the same word length. In non-linear quantizing, the data subject to quantizing are pre-processed by a non-linear function, and the resulting pre-processed data are then represents a linear quantizing of x, and f(x) is a non-linear function of technique is specified by the ISO/MPEG layer three audio coding standard, for example. quantizing levels. FIG. 1 shows the relationship between the data value x of the data subject to quantizing and the resulting quantizing levels. The five possible quantizing levels denoted by the circled integers above the horizontal axis. Each quantizing level has a corresponding quantizing value. For example, in FIG. 1, the quantizing level 2 has a quantizing value of 1.0, and the quantizing level -1 has a quantizing value of -0.5. The data value x of the data subject to quantizing and the quantizing level of the quantizing values are indicated in FIG. 1 by the values -1.0, -0.5, 0, 0.5, and 1.0 below the horizontal axis.
In the linear quantizing scheme illustrated in FIG. 1, the data value x is quantized to the quantizing level having a quantizing value that is closest to x. In FIG. 1, the decision values between adjacent quantizing values are indicated by broken lines. For example, the broken line D.sub.01 indicates the decision
REFERENCES:
patent: 3973081 (1976-08-01), Hutchins
patent: 4202018 (1980-05-01), Stockham, Jr.
patent: 4386237 (1983-05-01), Virupaksha
patent: 4516241 (1985-05-01), Farah et al.
patent: 4713776 (1987-12-01), Araseki
patent: 4809274 (1989-02-01), Walker et al.
patent: 4870685 (1989-09-01), Kadokawa et al.
patent: 5134475 (1992-07-01), Johnston et al.
patent: 5150387 (1992-09-01), Yoshikawa et al.
patent: 5157760 (1992-10-01), Akagiri
patent: 5166686 (1992-11-01), Sugiyama
patent: 5166981 (1992-11-01), Iwahashi et al.
patent: 5185800 (1993-02-01), Mahieux
patent: 5197087 (1993-03-01), Iwahashi et al.
patent: 5285476 (1994-02-01), Akagiri et al.
patent: 5341457 (1994-08-01), Hall, II et al.
patent: 5375189 (1994-12-01), Tsutsui
patent: 5521713 (1996-05-01), Oikawa
patent: 5608713 (1997-03-01), Akagiri et al.
patent: 5617219 (1997-04-01), Oikawa
patent: 5623577 (1997-04-01), Fielder
patent: 5632003 (1997-05-01), Davidson et al.
patent: 5634082 (1997-05-01), Shimoyoshi et al.
patent: 5642383 (1997-06-01), Suzuki
Roger C. Wood, "On Optimum Quantization," IEEE Transactions on Information Theory, vol. IT-15, No. 2, pp. 248-252, Mar. 1969.
Joel Max, "Quantizing for Minimum Distortion," IRE Transactions on Information Theory, vol. IT-6, No. 1, pp. 7-12, Mar. 1960.
`A Unified Formulation of Segment Companding Laws and Synthesis of Codecs and Digital Compandors`, H. Kaneko Bell System Technical Journal, vol. 49, No. 7, Sep. 1970 New York, U.S. pp. 1555-1588.
Digitally Linearizable Compandors with Comments on "Project for a Digital Telephone Network"Wm.T. Montgomery--IEEE Transactions on Communication Technology, vol. com18, No. 1, Feb. 1970 New York, U.S.
Hudspeth David R.
Smits Talivaldis Ivars
LandOfFree
Methods for non-linearly quantizing and non-linearly dequantizin does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Methods for non-linearly quantizing and non-linearly dequantizin, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Methods for non-linearly quantizing and non-linearly dequantizin will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2428813