Word spotting false alarm phrases
View Patent ↗In one aspect, a method for processing media includes accepting a query. One or more language patterns are identified that are similar to the query. A putative instance of the query is located in the media. The putative instance is associated with a corresponding location in the media. The media in a vicinity of the putative instance is compared to the identified language patterns and data characterizing the putative instance of the query is provided according to the comparing of the media to the language patterns, for example, as a score for the putative instance that is determined according to the comparing of the media to the language patterns.
1. A computer implemented method for processing media comprising:
using a computer to perform the steps of:
accepting the media;
accepting language data;
accepting a query;
processing the query using the language data to identify one or more language patterns each having a length larger than a length of a language pattern of the query, each language pattern including a first portion that is acoustically similar to the language pattern of the query and a second portion that is not acoustically similar to the language pattern of the query;
locating a putative instance of the query in the media, the putative instance being associated with a first time interval in the media, the locating including determining a match score for the located putative instance of the query to the media in the first time interval;
selecting a second time interval of the media according to the first time interval, the second time interval being larger than and including the first time interval;
determining a degree of acoustic similarity between the media in the second time interval and one or more of the identified language patterns;
determining a score for the putative instance according to the match score for the located putative instance of the query and to at least one score associated with the determined degree of acoustic similarity between the media in the second time interval and one or more of the identified language patterns; and
providing data characterizing the putative instance of the query according to the determined degree of acoustic similarity, including identifying the putative instance of the query as a potential falsely identified instance of the query based on the determined score for the putative instance.
2. The method of claim 1 wherein providing data characterizing the putative instance includes providing data representing a visual display of the putative instance and at least some of the language patterns.
3. The method of claim 1 wherein determining the degree of acoustic similarity between the media in the second time interval and one or more of the identified language patterns includes determining a match score for each of the language patterns to the media in the second time interval.
4. The method of claim 3 wherein determining the score for the putative instance according to the determined degree of acoustic similarity includes combining the match score for the query and match scores for one or more of the language patterns.
5. The method of claim 4 wherein processing the query using the language data to identify one or more language patterns that each include the first portion of the pattern that is acoustically similar to the query includes accepting a specification of at least some of said patterns with a specification of the query.
6. The method of claim 4 wherein processing the query using the language data to identify one or more language patterns that each include the first portion of the pattern that is acoustically similar to the query includes identifying at least some of said patterns as phonetically similar to at least a portion of the query.
7. The method of claim 1 wherein processing of the query using the language data to identify one or more language patterns that each include the first portion of the pattern that is acoustically similar to the query is performed prior to locating the putative instance of the query.
8. The method of claim 7 wherein the method further comprises presenting the one or more language patterns to a user from whom the query is accepted.
9. The method of claim 8 wherein presenting the one or more patterns to the user includes determining a degree of similarity of each of the patterns to the query.
10. The method of claim 7 wherein determining the degree of acoustic similarity between the media in the second time interval and one or more of the identified language patterns is performed after locating the putative instance.
11. The method of claim 1 wherein providing data characterizing the putative instance of the query according to determined degree of acoustic similarity includes providing data representing the putative instance and at least some of the language patterns, and the method further comprises accepting an indicator that one of the language patterns represents the media in the location of the query.
12. The method of claim 11 wherein the method further comprises adapting a configuration of a wordspotting engine according to the accepted indicator to improve an accuracy of locating further putative instances of the query.
13. The method of claim 1 wherein processing the query using the language data to identify one or more language patterns that each include the first portion of the pattern that is acoustically similar to the query includes accepting a specification of at least some of said patterns with a specification of the query.
14. The method of claim 1 wherein processing the query using the language data to identify one or more language patterns that each include the first portion of the pattern that is acoustically similar to the query includes combining word information and phoneme substitution information with a phonetic representation of the query.
15. The method of claim 1 wherein processing the query using the language data to identify one or more language patterns that each include the first portion of the pattern that is acoustically similar to the query includes identifying language patterns that include a portion that is phonetically similar to the query.
16. The method of claim 1 wherein processing the query using the language data to identify one or more language patterns that each include the first portion of the pattern that is acoustically similar to the query includes identifying one or more word sequences.
17. The method of claim 1 , wherein processing the query using the language data to identify one or more language patterns that each include the first portion of the pattern that is acoustically similar to the query includes identifying at least one language pattern such that the query is similar to a portion less than the entire language pattern.
18. The method of claim 1 , wherein the one or more language patterns phonetically similar to the query are represented as a finite state transducer.
19. The method of claim 1 , wherein processing the query using the language data to identify one or more language patterns that each include the first portion of the pattern that is acoustically similar to the query comprises selecting N best language patterns from representation of a larger plurality of language patterns.
20. The method of claim 19 , wherein the representation of the larger plurality of language patterns comprises a finite-state transducer and selecting the N best language patterns comprises identifying multiple paths through the transducer.
21. The method of claim 1 wherein the query includes a sequence of one or more words having a phonetic representation and each identified language pattern of the one or more language patterns includes a part that is phonetically similar to the query.
22. The method of claim 1 wherein each of the one or more language patterns includes a first plurality of phonemes, the language pattern of the query includes a second plurality of phonemes, and a number of phonemes in the first plurality of phonemes is larger than a number of phonemes in the second plurality of phonemes.
23. Software stored on non-transitory computer-readable media comprising:
instructions for causing a data processing system to:
accept media;
accept language data;
accept a query;
process the query using the language data to identify one or more language patterns each having a length larger than a length of a language pattern of the query, each language pattern including a first portion that is acoustically similar to the language pattern of the query and a second portion that is not acoustically similar to the language pattern of the query;
locate a putative instance of the query in the media, the putative instance being associated with a first time interval in the media, the locating including determining a match score for the located putative instance of the query to the media in the first time interval;
select a second time interval of the media according to the first time interval, the second time interval being larger than and including the first time interval;
determine a degree of acoustic similarity between the media in the second time interval and one or more of the identified language patterns;
determine a score for the putative instance according to the match score for the located putative instance of the query and to at least one score associated with the determined degree of acoustic similarity between the media in the second time interval and one or more of the identified language patterns; and
provide data characterizing the putative instance of the query according to the determined degree of acoustic similarity, including identifying the putative instance of the query as a potential falsely identified instance of the query based on the determined score for the putative instance.
24. A media processing system comprising:
an input for accepting media;
an input for accepting language data;
and input for accepting a query;
a computer implemented module for processing the query using the language data to identify one or more language patterns each having a length larger than a length of a language pattern of the query, each language pattern including a first portion that is acoustically similar to the language pattern of the query and a second portion that is not acoustically similar to the language pattern of the query;
a computer implemented word spotting module including a first word spotting component for accepting media and locating a putative instance of the query in the media, the putative instance being associated with a first time interval in the media, the locating including determining a match score for the located putative instance of the query to the media in the first time interval; and
a computer implemented second component for selecting a second time interval of the media according to the first time interval, the second time interval being larger than and including the first time interval; determining a degree of acoustic similarity between the media in the second time interval and one or more of the identified language patterns; and determining a score for the putative instance according to the match score for the located putative instance of the query and to at least one score associated with the determined degree of acoustic similarity between the media in the second time interval and one or more of the identified language patterns; and
a computer implemented third component for providing data characterizing the putative instance of the query according to the determined degree of acoustic similarity, including identifying the putative instance of the query as a potential falsely identified instance of the query based on the determined score for the putative instance.