System and method for fingerprinting datasets
Systems and methods for the matching of datasets, such as input audio segments, with known datasets in a database are disclosed. In an illustrative embodiment, the use of the presently disclosed systems and methods is described in conjunction with recognizing known network message recordings encountered during an outbound telephone call. The methodologies include creation of a ternary fingerprint bitmap to make the comparison process more efficient. Also disclosed are automated methodologies for creating the database of known datasets from a larger collection of datasets.
1. A method for extracting acoustic fingerprints for common recordings comprising the steps of:
a. identifying segments of interest from a dataset of recordings by a means of classification using a Call Progress Analysis system;
b. generating acoustic fingerprint candidates for said identified segments, each of the acoustic fingerprint candidates being computed from a plurality of power spectrums of the corresponding one of the identified segments, each of the power spectrums comprising a plurality of frequency bands;
c. adding said acoustic fingerprint candidates to an acoustic fingerprint database;
d. repeating steps (a) and (b) to process the recordings in the dataset using the acoustic fingerprint database containing said acoustic fingerprint candidates added in step (c); and
e. removing said acoustic fingerprint candidates from the acoustic fingerprint database that were not detected a specified number of times.
2. The method of claim 1 , wherein said segments of interest comprise audio segments that do not comprise any of: silence, simple tones, background noise, and known fingerprints.
3. The method of claim 1 , wherein the specified number of times is 2.
4. A method for extracting fingerprints for common data comprising the steps of:
a. identifying subsets of interest from a dataset by means of classification using a Call Progress Analysis system;
b. generating fingerprint candidates for said subsets, each of the fingerprint candidates being computed from a plurality of power spectrums of the corresponding one of the identified segments, each of the power spectrums comprising a plurality of frequency bands;
c. adding said fingerprint candidates to a fingerprint database;
d. repeating steps (a) and (b), to process the dataset using the fingerprint database in step (c) containing the additions of fingerprint candidates; and
e. removing said fingerprint candidates from the database that do not meet a specified criteria.
5. The method of claim 4 , wherein said common data comprises audio recordings.
6. The method of claim 4 , wherein said subsets of interest comprise audio subsets that do not comprise any of: simple tones, background noise, and known fingerprints.
7. The method of claim 4 , wherein said criteria is the number of times the fingerprint matches.
8. The method of claim 7 , wherein said number of times is 2.