Pattern distance calculating equipment

• US 4,601,054 A
• Filed: 07/23/1985
• Issued: 07/15/1986
• Est. Priority Date: 11/06/1981
• Status: Expired due to Term

• Current Assignee: Nippon Electric Company Limited (NEC Corporation)
• Original Assignee: Nippon Electric Company Limited (NEC Corporation)
• Inventors: Watari, Masao, Chiba, Seibi
• Primary Examiner(s): Boudreau, Leo H.
• Application Number:
  US06/757,929
• Time in Patent Office: 357 Days
• Field of Search: 382/1, 382/34, 382/36, 381/40-50
• US Class Current: 704/238
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates

Abstract

A pattern recognition apparatus for recognizing spoken words of a nonspecific speaker or of a specific speaker. A reference pattern composed of a sequence of feature vectors, each composed of n feature parameters, b_i, is stored. The reference pattern represents a form of average of said words to be recognized as determined by multiple reference speakers speaking the same words or by the specified speaker speaking said words several times. A deviation pattern composed of a sequence of feature vectors composed of n feature parameters, w_i /2, is stored. The deviation pattern is a measure of the deviation from the reference of the repeated utterances of the reference speakers or the specified speaker. An input pattern, representing the utterances of a speaker, is composed of a sequence of feature vectors, each composed of n feature parameters, a_i, and is stored. A measure of the similarity of the input and reference patterns is calculated, taking into account the deviation pattern, according to one of several possible distance formulae. Basically, a distance parameter calculated for each corresponding input, reference and deviation parameter is set to zero value if the input parameter is inside the deviation range of the reference parameter, and is otherwise calculated to be a finite value.

Claims

1. A pattern distance calculating apparatus comprising:

an input pattern memory storing an input pattern comprised of a sequence of feature vectors each composed of a predetermined number n (i.e., a positive integer) of feature parameters;

a reference pattern memory storing a reference pattern comprised of a sequence of feature vectors each composed of n feature parameters of the same type as that of said input pattern;

a deviation pattern memory storing a deviation pattern indicative of a predetermined fluctuation range for each of said n feature parameters of said reference pattern; and

distance calculating means for calculating a distance between said input pattern and said reference pattern, said distance calculating means comprising;

first means for determining if the difference between the value of each feature parameter of said input pattern and the value of the corresponding feature parameter of said reference pattern is within the fluctuation range for said feature parameter of said reference pattern,

second means for selecting the value zero as a distance value if said difference is within said fluctuation range and for selecting a distance value determined by a preselected distance calculating formula if said difference is outside said fluctuation range, and

means for accumulating said distance values selected for each parameter to provide said distance between said input pattern and said reference pattern.

2. A pattern distance calculating apparatus as claimed in claim 1 for use as a specified speaker recognition apparatus wherein said reference pattern and said deviation pattern represent patterns obtained from a specified speaker uttering a given word or words several times, said reference pattern representing an average pattern of said speaker'"'"'s utterances and said deviation pattern representing the variations in the speaker'"'"'s utterances, and wherein said input pattern represents an input utterance of said specified speaker.

3. A pattern distance calculating apparatus as claimed in claim 1 for use as an unspecified speaker recognition apparatus wherein said reference pattern and said deviation pattern represent patterns obtained from several speakers uttering a given word or words, said reference pattern representing an average pattern of said speaker'"'"'s utterances and said deviation pattern representing the variations in the speaker'"'"'s utterances, and wherein said input pattern represents an input utterance of an unknown speaker.

4. A pattern distance calculating apparatus according to claim 1, wherein said means for selecting another finite non-zero value comprises means for calculating and selecting a Chebychev distance between said input and reference feature parameters.

5. A pattern distance calculating apparatus according to claim 1, wherein said means for selecting another finite non-zero value comprises means for calculating and selecting a Euclid distance between said input and reference feature parameters.

6. A pattern distance calculating apparatus according to claim 1, wherein said means for selecting another finite non-zero value comprises means for calculating and selecting a correlated distance between said input and reference feature parameters.

7. A pattern distance calculating apparatus according to claim 1, wherein each of the feature parameters of said reference pattern is an average of a plurality of corresponding feature parameters included in a plurality of predetermined refernce patterns and each of said predetermined fluctuation range is the standard deviation of a plurality of corresponding feature parameters included in a plurality of said predetermined reference patterns.

8. A pattern distance calculating apparatus according to claim 1, wherein each of the feature parameters of said reference pattern is a center value of a plurality of corresponding feature parameters included in a plurality of predetermined reference patterns and each of said predetermined fluctuation range is the subtracted value of the maximum minus the minimum of a plurality of corresponding feature parameters included in a plurality of said predetermined reference patterns.

9. A pattern distance calculating apparatus according to claim 1, said distance calculating means comprising:

a first subtractor means for subtracting a reference feature parameter output of said reference pattern memory from an input feature parameter output of said input pattern memory;

a first adder means for adding the output of said first subtractor and the deviation parameter corresponding to said reference feature parameter output of said deviation pattern memory;

a second subtractor means for subtracting the output of said first subtractor means from said deviation parameter;

first switching means for selectively switching to its output terminal the output of said first adder means, when the output of said first subtractor means is negative, and the output of said second subtractor means when the output of said first subtractor means is zero or positive;

second switching means for selectively switching to its output a signal indicative of zero, when the output of said first switching means is zero or positive, and a signal indicative of the output of said first switching means, when the output of said first switching means is negative;

a third subtractor means for subtracting the output of said second switching means applied to a first input terminal from a signal applied to a second input terminal thereof; and

an accumulator means for storing the output of said third subtractor means in an accumulated manner and for feeding its stored content to the second input of said third subtractor means.

10. A pattern distance calculating apparatus according to claim 1, said distance calculating means comprising:

a first subtractor means for subtracting a reference feature parameter output of said reference pattern memory from an input feature parameter output of said input pattern memory;

a first adder means for adding the output of said first subtractor and the deviation parameter corresponding to said reference feature parameter output of said deviation pattern memory;

a second subtractor means for subtracting the output of said first subtractor means from said deviation parameter;

first switching means for selectively switching to its output terminal the output of said first adder means, when the output of said first subtractor means is negative, and the output of said second subtractor means when the output of said first subtractor means is zero or positive;

second switching means for selectively switching to its output a signal indicative of zero, when the output of said first switching means is zero or positive, and a signal indicative of the output of said first switching means, when the output of said first switching means is negative;

a multiplier means for squaring the output of said second switching means;

a second adder means for adding the output of said multiplier means, applied to a first input thereof, and an input applied to a second input terminal thereof; and

an accumulator means for storing the output of said second adder means in an accumulated manner and for feeding its stored value to the second input of said second adder means.

11. A pattern distance calculating apparatus according to claim 1, said distance calculating means comprising:

a first subtractor means for substracting a reference feature parameter output of said reference pattern memory from an input feature parameter output of said input pattern memory;

a first adder means for adding the output of said first subtractor means and the deviation parameter output of said deviation pattern memory which deviation pattern corresponds to said reference feature parameter;

a second subtractor means for subtracting the output of said first subtractor means from said deviation parameter;

first switching means for selectively switching to an output terminal the output of said first adder means, when the output of said first subtractor means is negative, and the output of said second subtractor means when the output of said first subtractor means is zero or positive;

a second adder means for adding said input feature parameter output of said input pattern memory and said deviation parameter output of said deviation pattern memory;

a third subtractor means for subtracting said deviation parameter from said input feature parameter;

second switching means for selectively switching to an output thereof the output of said third subtractor means when the output of said first subtractor means is zero or positive, and the output of said second adder means when the output of said first subtractor means is negative;

third switching means for selectively switching to an output thereof said reference feature parameter output of said reference pattern memory when the output of said first switching means is zero or positive, and the output of said second switching means when the output of said first switching means is negative;

a multiplier means for multiplying the output of said third switching means and said reference feature parameter output of said reference pattern memory;

a third adder means for adding the output of said multiplier means, applied to a first input thereof, and a signal applied to a second input thereof; and

an accumulator means for storing the output of said third adder means in an accumulated manner and for feeding its stored value to the second input of said third adder.

12. A pattern distance calculating system in which the similarity between an input pattern composed of a sequence of feature vectors composed of a predetermined number n (i.e., a positive integer) of feature parameters and a reference pattern composed of a sequence of feature vectors composed of n feature parameters of the same type as that of said input pattern is calculated based on predetermined distance formulae including the respective feature parameters of said input pattern and the respective parameters of said reference pattern, wherein said distance is determined according to a deviation parameter representive of a predetermined fluctuation range for each of the parameters of said reference pattern such that the distance between said feature parameters of said input and reference patterns is set at zero or a predetermined value, when the value of the feature parameter of said input pattern at a certain time is within said fluctuation range of said reference pattern at a corresponding time, and at a value given by said predetermined formulae when said value is outside said fluctuation range.