Adaptive system for broadcast program identification and reporting

• US 5,581,658 A
• Filed: 12/14/1993
• Issued: 12/03/1996
• Est. Priority Date: 12/14/1993
• Status: Expired due to Fees

• Current Assignee: Fuzzy Logic, Infobase Systems Incorporated, Nadja K O'Hagan
• Original Assignee: Infobase Systems Incorporated
• Inventors: O'Hagan, Michael, O'Hagan, Nadia K., Hines, Frank E.
• Primary Examiner(s): Davis, George B.
• Application Number:
  US08/167,646
• Time in Patent Office: 1,085 Days
• Field of Search: 395/22, 395/24, 395/2.41
• US Class Current: 706/20
• CPC Class Codes: G06F 2218/12 Classification; Matching
• CPC Class Codes: H04H 20/14 for monitoring programmes
• CPC Class Codes: H04H 2201/90 characterised by the use of signatures
• CPC Class Codes: H04H 60/58 of audio determination or detection of speech characteristics in general G10L25/00; speech recognition in general G10L15/00

Abstract

A computer-implemented method and system for monitoring, identifying, classifying and logging musical work performance broadcasts over the public airwaves. The system uses a neural network to classify specially-processed "retinal" signatures of the musical work performance. The neural network is trained for each musical work using a single noise-biased retinal sample of the spectral distribution of preselected dynamic features of the corresponding audio signal. A detection decision is made at the neural network output using fuzzy logic circuitry to compare results of predetermined thresholding. The system of this invention fully automates the real-time identification of broadcast musical work performances.

Claims

1. A computer-implemented method for identifying and reporting an audible performance of a musical work recorded in a medium of fixed expression, said method comprising the steps of:

(a) training a neural network to generate an output signal corresponding to recognition of a selected retinal sample of said audible performance;

(b) receiving an audio signal from a performance broadcast;

(c) processing said audio signal to create a series of overlapping retinal samples of said audible performance, wherein each said retinal sample includes a plurality of digital signals representing a spectral and temporal distribution of power of said audio signal;

(d) generating said neural network output signal responsive to said series of overlapping retinal samples;

(e) comparing each said neural network output signal with a corresponding predetermined threshold signal; and

(f) logging a record of the broadcast of said musical work performance responsive to said neural network output signal exceeding said corresponding predetermined threshold signal.

2. The method of claim 1 wherein said retinal samples include selected portions of said audio signal from which invariant features are removed.

3. The method of claim 1 wherein said training step (a) is performed with said selected retinal sample in combination with random noise.

4. A computer-implemented method for identifying an I^th one of a plurality S of musical works from a first audio signal A_1s representing a first audible performance of a musical work, wherein for each of said plurality S of musical works a set of R weights (W_r)_s is predetermined responsive to a portion P_2s spanning a retinal time interval t_R of a second audio signal A_2s representing a second audible performance of said each musical work, wherein 1≦

s≦

S and 1≦

r≦

R are non-zero positive integers, said method comprising the steps of;

(a) creating a plurality R of retinal signals (S_r) representing a spectral and temporal distribution of power in a portion P_1s of said first audio signal A_1s spanning a retinal time interval t_R ;

(b) combining said retinal signals (S_r) with each of a plurality S of weight sets (W_r)_s to create a plurality S of neural output signals (N_s) each representing a weighted combination of said retinal signals (S_r) ;

(c) creating a plurality S of ratio signals (N_s /T_s) each corresponding to the ratio of each of said neural output signals (N_s) to a corresponding predetermined alarm threshold signal T_s ;

(d) if at least one of said ratio signals (N_s /T_s) exceeds unity, selecting an I^th neural output signal N_I =N_s corresponding to a greatest said ratio signal N_I /T_I of said plurality S of ratio signals (N_s /T_s), whereby said neural output signal N_I represents an identity of said I^th musical work, otherwise selecting nothing;

(e) making an alarm log record responsive to said selection of said neural output signal N_I, said alarm log record including an alarm date, an alarm time, and the identity of said I^th musical work; and

(f) repeating said steps (a)-(f) responsive to a passage of an updating time interval t_U <

t_R.

5. The method of claim 4 wherein said creating a plurality of R of retinal signals (Sr) comprises the steps of:

creating a plurality of first digital signals representing amplitude samples of said first audio signal portion P_1s ;

combining said first digital signals to create a set of second digital signals representing the Fourier Transform of said first digital signals spanning a predetermined frequency region; and

accumulating one or more said set of second digital signals to form said plurality R of retinal signals (S_r).

6. The method of claim 5 wherein said each weight set (w_r)_s is determined responsive to said second audio signal portion P_2s in combination with a random noise signal.

7. The method of claim 6 wherein each of said neural output signals (N_s) is exponentially smoothed by combining said each of said neural output signals (N_s) with a weighted combination of one or more earlier values of neural output signals (N_s) to create a smoothed neural output signal N_s.

8. The method of claim 7 wherein said first audible performance of said each musical work is included in a broadcast program.

9. The method of claim 4 wherein said first and second audible performances of said each musical work are the same performance recorded in a medium of fixed expression.

10. The method of claim 9 wherein said first audible performance of said each musical work is included in a broadcast program.

11. A system for identifying and logging a broadcast of an audible performance of a musical work recorded in a medium of fixed expression, said system comprising:

input means for receiving an audio signal from a performance broadcast;

processing means coupled to said input means for creating a series of overlapping retinal samples of audio signal, wherein each said retinal sample includes a plurality of digital signals representing a spectral and temporal distribution of power of said audio signal power;

neural network means coupled to said processing means for creating a neural output signal representing a recognition of said each retinal sample;

training means coupled to said neural network means for training said neural network means to generate said neural output signal corresponding to recognition of a selected retinal sample of said audible performance; and

logging means for storing a record of a recognition of said broadcast of said musical work performance responsive to each said neural output signal in excess of a predetermined threshold signal.

12. The system of claim 11 wherein said retinal samples include selected portions of said audio signal from which invariant features are removed.

13. The system of claim 11 wherein said training means further comprises:

noise biasing means for adding random noise to said selected retinal sample of said audible performance during said training of said neural network means.