Data processing: speech signal processing – linguistics – language – Speech signal processing – For storage or transmission
Reexamination Certificate
1996-09-30
2001-02-20
Zele, Krista (Department: 2748)
Data processing: speech signal processing, linguistics, language
Speech signal processing
For storage or transmission
Reexamination Certificate
active
06192336
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to lossy coding systems, and more particularly to vector quantization for use in codebook-based coding systems.
BACKGROUND OF THE INVENTION
Digital speech processing typically can serve several purposes in computers. In some systems, speech signals are merely stored and transmitted. Other systems employ processing that enhances speech signals to improve the quality and intelligibility. Further, speech processing is often utilized to generate or synthesize waveforms to resemble speech, to provide verification of a speaker's identity, and/or to translate speech inputs into written outputs.
In some speech processing systems, speech coding is performed to reduce the amount of data required for signal representation, often with analysis by synthesis adaptive predictive coders, including various versions of vector or code-excited coders. In the predictive systems, models of the vocal cord shape, i.e., the spectral envelope, and the periodic vibrations of the vocal cord, i.e., the spectral fine structure of speech signals, are typically utilized and efficiently performed through slowly, time-varying linear prediction filters.
These models typically utilize parameters to replicate as closely as possible the original speech signal. There tends to be numerous parameters involved in such modeling. Compression schemes are often employed to reduce the number of parameters requiring transmission in the processing system. One such technique is known as vector quantization.
Generally, vector quantization schemes, whether in speech processing or other large data modeling systems, such as image processing systems, employ a codebook or vocabulary of codevectors, and an index to the codebook. An optimal codevector in the codebook relative to an input vector is usually determined through intensive computations. An index to the optimal codevector is then transmitted. Thus, vector quantization effectively reduces the amount of information transmitted by transmitting only an indexed reference to the codevector, rather than the entire codevector. Unfortunately, the intensive computations typically involved in the determination of the optimal codevector are time-consuming and thus expensive. Accordingly, a need exists for a more efficient search strategy in vector quantization.
SUMMARY OF THE INVENTION
The present invention addresses such a need and provides method and system aspects for searching for an optimal codevector from a plurality of codevectors in a codebook, the optimal codevector having a minimum weighted distance to a given vector. The preferred aspects determine a partial distance with a current vector component of a current codevector and the corresponding component of the given vector, compare the partial distance to a saved “renormalized” minimum partial distance, and proceed to a next codevector when the saved renormalized minimum partial distance is smaller than the partial distance.
In addition, the present invention proceeds to a next vector component when the partial distance is smaller than the corresponding saved renormalized minimum partial distance. When the partial distance computed with each next vector component is smaller than the saved renormalized minimum partial distance, the present invention calculates a full weighted distance value. Further, the weighted full distance is compared to a saved minimum full weighted distance, and when the full weighted distance is smaller than the saved minimum full weighted distance, an optimal index to the current codevector is updated, the saved minimum full weighted distance is updated to the full weighted distance, and the new renormalized minimum partial distances are determined and stored. The operation then continues with a next codevector until all codevectors have been used. An optimal index to identify the optimal codevector is then returned when all codevectors in the codebook have been used.
With the present invention, an advantageous computation of partial distance values reduces the computational load of searching for an optimal codevector in a given codebook. Further, the present invention achieves such advantage without restricting the codebook to a predefined structure or performing preliminary processing to determine a structure for the codebook. These and other advantages of the aspects of the present invention will be more fully understood in conjunction with the following detailed description and accompanying drawings.
REFERENCES:
patent: 3979557 (1976-09-01), Schulman et al.
patent: 4944013 (1990-07-01), Gouvianakis
patent: 5010574 (1991-04-01), Wang
patent: 5091945 (1992-02-01), Kleijn
patent: 5127053 (1992-06-01), Koch
patent: 5142584 (1992-08-01), Ozawa
patent: 5195137 (1993-03-01), Swaminathan
patent: 5216747 (1993-06-01), Hardwick et al.
patent: 5230036 (1993-07-01), Akamine
patent: 5271089 (1993-12-01), Ozawa
patent: 5323486 (1994-06-01), Taniguchi
patent: 5377301 (1994-12-01), Rosenberg et al.
patent: 5455888 (1995-10-01), Iyengar
patent: 5481739 (1996-01-01), Staats
patent: 5487128 (1996-01-01), Ozawa
patent: 5491772 (1996-02-01), Hardwick et al.
patent: 5574823 (1996-11-01), Hassanein
patent: 5596676 (1997-01-01), Swaminathan
patent: 5642464 (1997-06-01), Yue
patent: 5664055 (1997-09-01), Kroon
Chu Ke-Chiang
Manduchi Roberto
Ponceleon Dulce
Wu Hsi-Jung
Apple Computer Inc.
Opsasnick Michael N.
Sawyer Law Group LLP
Zele Krista
LandOfFree
Method and system for searching for an optimal codevector 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 and system for searching for an optimal codevector, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and system for searching for an optimal codevector will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2611851