Pattern recognition system and method

• US 5,778,342 A
• Filed: 02/01/1996
• Issued: 07/07/1998
• Est. Priority Date: 02/01/1996
• Status: Expired due to Term

• Current Assignee: Wireless 3G
• Original Assignee: DSPC Israel Limited
• Inventors: Erell, Adoram, Burshtein, David
• Primary Examiner(s): Hafiz, Tariq R.
• Application Number:
  US08/595,357
• Time in Patent Office: 887 Days
• Field of Search: 395/2.4, 395/2.45, 395/2.5, 395/2.52, 395/2.53, 395/2.48, 704/200, 704/231, 704/236, 704/239, 704/241-247
• US Class Current: 704/256.8
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/12 using dynamic programming techniques, e.g. dynamic time warping [DTW]
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 21/0216 characterised by the method used for estimating noise

Abstract

A pattern recognition system and method is disclosed. The method includes the steps of a) providing a noisy test feature set of the input signal, a plurality of reference feature sets of reference templates produced in a quiet environment, and a background noise feature set of background noise present in the input signal, b) producing adapted reference templates from the test feature set, the background noise feature set and the reference feature sets and c) determining match scores defining the match between each of the adapted reference templates and the test feature set. The method can also include adapting the scores before accepting a score as the result. The system and method are described for both Hidden Markov Model (HMM) and Dynamic Time Warping (DTW) scoring units. The system performs the steps of the method.

Claims

1. A pattern recognition system comprising:

means for providing a test feature set of a generally noisy input signal characterizing at least a portion of an input pattern contained within said input signal;

means for providing a plurality of reference feature sets of reference templates produced in a quiet environment;

means for providing a background noise feature set of background noise present in said input signal;

a template adapter for producing adapted reference templates from said test feature set, said background noise feature set and said reference feature sets; and

a global scoring unit for determining match scores defining the match between each of said adapted reference templates and said test feature set,

wherein said feature sets are autocorrelation feature sets and said template adapter includes;

means for raising the gain level of a reference feature set to the value of the difference of the average energy of said test feature set and the average energy of said background noise feature set; and

means for adjusting said gain-raised reference feature set by adding to it said background noise feature set thereby to create said adapted reference templates.

2. A pattern recognition system according to claim 1 and also including a signal to noise ratio determiner of the signal to noise ratio in the input signal and a decision unit for accepting at least one of said match scores if the signal to noise ratio and score value are within predetermined values.

3. A pattern recognition system according to claim 2 wherein said global scoring unit is a Dynamic Time Warping (DTW) global scoring unit.

4. A pattern recognition system according to claim 3 wherein said DTW global scoring unit operates on cepstral representations of said autocorrelation feature sets.

5. A pattern recognition system according to claim 1 wherein said global scoring unit is a Hidden Markov Model (HMM) global scoring unit.

6. A pattern recognition system according to claim 5 wherein said features sets are autocorrelation feature sets for the current frame and for another frame and wherein said HMM global scoring unit operates on cepstral representations of said feature sets for the current frame and on a cepstral difference between the cepstral representation of said feature sets for the current and another frame.

7. A pattern recognition system according to claim 6 and wherein said means for providing a plurality of reference feature sets includes:

means for producing cepstral representations of the autocorrelation feature sets of a plurality of frames of a plurality of reference signals and the cepstral difference between two frames of each reference signal;

means for determining the cepstral mean of a mixture of frames;

means for inverse-transforming the cepstral mean of said current frame into an autocorrelation vector for said current frame;

means for adding the cepstral mean of the current frame to said time difference and inverse-transforming the result thereby to produce an autocorrelation vector for said another frame; and

means for generating said autocorrelation feature set from said autocorrelation vector of said current and another frame.

8. A pattern recognition system according to claim 7 wherein said adapted reference templates includes adapted reference templates of said current and another frame and wherein said template adapter includes means for generating an adapted cepstral difference from the cepstral representations of said adapted reference templates of said current and another frame.

9. A pattern recognition system according to claim 5 and also including a decision unit for accepting the best one of said match scores in accordance with a predetermined criterion.

10. A method for pattern recognition, the method comprising the steps of:

providing a test feature set of a generally noisy input signal characterizing at least a portion of an input pattern contained within said input signal;

providing a plurality of reference feature sets of reference templates produced in a quiet environment;

providing a background noise feature set of background noise present in said input signal;

producing adapted reference templates from said test feature set, said background noise feature set and said reference feature sets; and

determining match scores defining the match between each of said adapted reference templates and said test feature set

wherein said feature sets are autocorrelation feature sets and said step of producing includes the steps of;

raising the gain level of a reference feature set to the value of the difference of the average energy of said test feature set and the average energy of said background noise feature set; and

adjusting said gain-raised reference feature set by adding to it said background noise feature set thereby to create said adapted reference templates.

11. A method according to claim 10 and also including the steps of determining a signal to noise ratio in the input signal and accepting at least one of said match scores if the signal to noise ratio and score value are within predetermined values.

12. A method according to claim 11 wherein said step of determining performs Dynamic Time Warping (DTW).

13. A method according to claim 12 wherein said step of determining operates on cepstral representations of said autocorrelation feature sets.

14. A method according to claim 10 wherein said step of determining performs Hidden Markov Model (HMM) scoring.

15. A method according to claim 14 wherein said features sets are autocorrelation feature sets for the current frame and for another frame and wherein said step of determining operates on the cepstral representations of said autocorrelation feature sets for the current frame and on a cepstral difference between the cepstral representations of said feature sets for the current and another frame.

16. A method according to claim 15 and wherein said step of providing a plurality of reference feature sets includes the steps of:

producing cepstral representations of the autocorrelation feature sets of a plurality of frames of a plurality of reference signals;

producing the cepstral difference between two frames of each reference signal;

determining the cepstral mean of a mixture of frames;

inverse-transforming the cepstral means of said current frame into an autocorrelation vector for said current frame;

adding the cepstral means of the current frame to said time difference and inverse-transforming the result thereby to produce an autocorrelation vector for said another frame; and

generating said autocorrelation feature set from said autocorrelation vector of said current and another frame.

17. A method according to claim 16 wherein said adapted reference templates includes adapted reference templates of said current and another frame and wherein said step of producing includes the steps of raising the gain level of a reference feature set to the value of the difference of the average energy of said test feature set and the average energy of said background noise feature set and adjusting said gain-raised reference feature set by said background noise feature set thereby to create said adapted reference templates of said current and includes the step of generating an adapted cepstral difference from the cepstral representations of said adapted reference templates of said current and another frame.

18. A method according to claim 17 and also including the step of accepting the best one of said match scores in accordance with a predetermined criterion.