Training circuit for audio signal recognition computer

• US 4,297,528 A
• Filed: 09/10/1979
• Issued: 10/27/1981
• Est. Priority Date: 09/10/1979
• Status: Expired due to Term

• Current Assignee: GTE Mobile Communications Service Corporation (Verizon Communications Inc.)
• Original Assignee: Interstate Electronics Corp. (L3Harris Technologies, Inc.)
• Inventors: Beno, Richard D.
• Primary Examiner(s): Nusbaum, Mark E.
• Assistant Examiner(s): Kemeny, E. S.
• Application Number:
  US06/073,781
• Time in Patent Office: 778 Days
• Field of Search: 179/1 SB, 179/1 SC, 179/1 SD, 179/1 SF, 340/146.3 T, 340/146.3 S, 340/146.3 Q
• US Class Current: 704/244
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 25/87 Detection of discrete points within a voice signal

Abstract

A signal pattern recognition system includes a reference pattern memory storing plural reference data patterns against which input patterns are compared for recognition. The reference data patterns are formed by merging training patterns together. Each training pattern, to be accepted for merging, must match the previously merged patterns by a threshold amount. The threshold is automatically varied as the number of previously merged training patterns increases. If a predetermined number of successive training patterns is below the threshold, the training process is repeated, from the beginning, for the reference pattern which generates the errors. The system automatically trains each reference pattern with the same number of training patterns to assure uniformity when input data patterns are compared for recognition.

Claims

1. Apparatus for signal pattern recognition, comprising:

means for storing a plurality of reference data patterns;

means for comparing a signal pattern with said reference data patterns for recognition; and

means for updating a selected one of said reference data patterns comprising;

means for inputting a training signal pattern;

means for comparing said training signal pattern with each said reference data pattern to generate a score signal for each said reference data pattern indicative of the similarities therebetween;

means for selecting the reference data pattern with the highest score for similarity with said training signal pattern;

means for combining said selected reference data pattern with said training pattern to produce a new reference data pattern; and

means for replacing said reference data pattern with said new reference data pattern in said storing means if said score signal is above a predetermined level without affecting any other said reference data patterns regardless of the scores of said other reference data patterns in comparison with said training signal pattern.

2. Apparatus for signal pattern recognition, as defined in claim 1, wherein said updating means further comprises:

means for adjusting said predetermined level.

3. Apparatus for signal pattern recognition, as defined in claim 1, wherein said updating means further comprises:

means for counting the number of sequential training signal patterns which produce score signals below a predetermined level when compared against said selected reference data pattern with the highest score for similarity; and

means responsive to said counting means for initiating additional training of said selected reference data pattern when the count reaches a predetermined number.

4. Apparatus for signal pattern recognition, comprising:

means for storing plural reference data patterns;

means for adjusting and comparing a training signal pattern with said plural reference data patterns for recognition;

means for combining a plurality of training signal patterns to produce each of said plural reference data patterns; and

means for controlling said inputting and combining means to combine the identical number of training signal patterns for each of said plural reference data patterns.

5. Apparatus for signal pattern recognition as defined in claim 4 wherein said controlling means comprises:

means for storing a number equal to the number of training signal patterns previously combined in said combining means to produce said reference data pattern;

means for storing a desired number of training signal patterns;

means for comparing said stored number of previously combined training signal patterns with said stored desired number of training signal patterns; and

means responsive to said comparing means for controlling said inputting and combining means.

6. Apparatus for signal pattern recognition comprising:

a memory storing multiple reference data patterns;

means for comparing said signal patterns against said multiple reference data patterns for recognition; and

a circuit for storing said multiple reference data patterns in said memory, comprising;

means for inputting and storing new signal patterns for cumulative use in generating one of said multiple reference data patterns;

means for combining said new signal patterns with said one reference data pattern previously stored in said memory to produce a new reference data pattern;

means for replacing said one reference data pattern previously stored with said new reference data pattern in said memory; and

means for sequencing said inputting and storing means to assure that said inputting, combining and merging means operate an identical number of times on each of said multiple reference data patterns in said memory.

7. A signal pattern encoder and classifier comprising:

a transducer;

signal processing means coupled to the output of said transducer for providing spectrum samples representative of the signal output of said transducer;

event encoding logic means coupled to the output of said signal processing means for computing the difference between a current spectrum sample and the last previous spectrum sample accepted by said event encoding logic means;

means responsive to said event encoding means for rejecting said current spectrum sample if said difference is below a predetermined value;

timing and control means coupled to said signal processing means and said event encoding logic means;

reference pattern memory means coupled with said timing and control means; and

pattern classification means connected to said event encoding means, said timing and control means and said reference pattern memory means wherein said pattern classification means comprises;

means for notifying the apparatus user to submit additional data;

means responsive to said notification means for causing samples to be taken of the additional data;

means responsive to incoming samples for comparison and coding for those samples;

storage means connected to said comparison and coding means for storing the coded data;

means connected to said reference pattern memory means for comparison of incoming data with said reference patterns in said reference pattern memory means and assigning a comparison value to said incoming data for each said comparison;

means responsive to said assigned comparison value for selecting the reference data pattern with the highest comparison value with said incoming data and for determining whether the assigned comparison value for said selected reference data pattern against said incoming data sample is acceptable; and

means responsive to said determining means and connected to said reference pattern memory means for storing said incoming data sample into said reference pattern memory means replacing only said selected reference data pattern with a new reference data pattern generated by combining said selected reference data pattern with said incoming data sample and leaving all other reference data patterns unaffected regardless of the comparison value assigned to any of said nonselected reference data patterns.

8. A signal pattern encoder, as defined in claim 7, further comprising:

second storage means retaining a count of the number of incoming samples combined to form each said reference data pattern; and

third storage means retaining a count of the number of sequential samples not meeting said predetermined threshold value for a selected reference data pattern.

9. A signal pattern encoder, as defined in claim 8, wherein said pattern classification means further comprises:

means responsive to said third storage means for discarding a selected reference data pattern when the count in said third storage means reaches a predetermined number for said selected reference data pattern.

10. A signal pattern encoder, as defined in claim 8, wherein said pattern classification means further comprises:

means responsive to said second storage means for varying said predetermined threshold in accordance with the count stored in said second means.

11. An apparatus for signal pattern recognition, comprising:

means for storing a reference data pattern;

means for comparing a signal pattern with said reference data pattern for recognition;

means for updating said reference data pattern comprising;

means for inputting a training signal pattern;

means for comparing said training signal pattern with said reference data pattern to generate a score signal indicative of the similarities therebetween;

means for combining said reference data pattern with said training pattern to produce a new reference data pattern; and

means for replacing said reference data pattern with said new reference data pattern in said storing means if said score signal is above a predetermined level; and

means for adjusting said predetermined level including;

means for counting the number of said training signal patterns which have been combined by said combining means to produce said reference data pattern; and

means for setting said predetermined level as a function of the output count of said counting means.

12. An apparatus for signal pattern recognition, as defined in claim 11, wherein said setting means increases said predetermined level in response to an increased count output from said counting means.

13. An apparatus for signal pattern recognition, as defined in claim 11, wherein said setting means comprises:

a read only memory.

14. Apparatus for signal pattern recognition, comprising:

means for storing a reference data pattern;

means for comparing a signal pattern with said reference pattern for recognition;

means for updating said reference data pattern comprising;

means for inputting a training signal pattern;

means for comparing said training signal pattern with said reference data pattern to generate a score signal indicative of the similarities therebetween;

means for combining said reference data pattern with said training pattern to produce a new reference data pattern;

means for replacing said reference data pattern with said new reference data pattern in said storing means if said score signal exceeds a predetermined level;

means for counting sequential training signal patterns which produce score signals from said comparing means below a predetermined level; and

means responsive to said counting means for initiating additional training of said reference data pattern; and

wherein said initiating means comprises;

means for storing a number equal to the number of training signal patterns previously combined in said combining means to produce said reference data pattern;

means for operating said training signal pattern inputting means in response to said number of previously combined training signals; and

means for setting said number of training signal patterns in said storing means to a lower value in response to said counting means.

15. Apparatus for signal pattern recognition, as defined in claim 14, wherein said setting means sets said number of training signal patterns in said storing means to zero.

16. A signal pattern encoder and classifier comprising:

a transducer;

signal processing means coupled to the output of said transducer for providing spectrum samples representative of the signal output of said transducer;

event encoding logic means coupled to the output of said signal processing means for computing the difference between a current spectrum sample and the last previous spectrum sample accepted by said event encoding logic means;

means responsive to said event encoding means for rejecting said current spectrum sample if said difference is below a predetermined value;

timing and control means coupled to said signal processing means and said event encoding logic means;

reference pattern memory means coupled with said timing and control means;

pattern classification means connected to said event encoding means, said timing and control means and said reference pattern memory means wherein said pattern classification means comprises;

means for notifying the apparatus user to submit additional data;

means responsive to said notification means for causing samples to be taken of the additional data;

means responsive to incoming samples for comparison and coding for those samples;

storage means connected to said comparison and coding means for storing the coded data;

means connected to said reference pattern memory means for comparison of incoming data and assigning a comparison value to said incoming data;

means responsive to said assigned comparison value for determining whether the particular coded incoming data sample is acceptable above a predetermined threshold value; and

means responsive to said determining means and connected to said reference pattern memory means for storing said incoming data sample into said reference pattern memory means; and

wherein said threshold value is varied automatically in response to the number of incoming samples used to generate the data in said reference pattern memory means.