Back to results
US 8,712,774 B2Granted
Systems and methods for generating a hybrid text string from two or more text strings generated by multiple automated speech recognition systems
AssigneeNuance Communications, Inc.
InventorJonathan Wiggs
Life of this patent
Estimating the term…
Abstract
A hybrid text generator is disclosed that generates a hybrid text string from multiple text strings that are produced from an audio input by multiple automated speech recognition systems. The hybrid text generator receives metadata that describes a time-location that each word from the multiple text strings is located in the audio input. The hybrid text generator matches words between the multiple text strings using the metadata and generates a hybrid text string that includes the matched words. The hybrid text generator utilizes confidence scores associated with words that do not match between the multiple text strings to determine whether to add an unmatched word to the hybrid text string.
Claims24
A method of creating a hybrid text string, wherein the method is performed by a computing system having a processor and a memory, the method comprising:
- receiving a first time-indexed text string generated from an audio input by a first automated speech recognition system;
- receiving a second time-indexed text string generated from the audio input by a second automated speech recognition system;
- identifying at least one anchor word, wherein the at least one anchor word is associated with an anchor word point,
- wherein the at least one anchor word is a word having a high probability of being correctly recognized by both the first and second automated speech recognition systems, and,
- wherein the anchor word point is located at a time location t1 in the first time-indexed text string and at a time location t2 in the second time-indexed text string, wherein time locations t1 and t2 are different;
- calculating a match vector, wherein the match vector is associated with a variance between the time location t1 in the first time-indexed text string and the time location t2 in the second time-indexed text string;
- identifying at least one matched word, wherein the at least one matched word is located at a time location t3 in the first time-indexed text string and at a time location t4 in the second time-indexed text string,
- wherein time location t4 is located within a second time range in the second time-indexed text string,
- wherein the second time range is generated by combining at least the time location t3, the match vector, and a margin of error;
- and
- creating a hybrid text string, wherein the hybrid text string includes the at least one anchor word and the at least one matched word, and wherein the hybrid text string includes at least one word from the first time-indexed text string that has a higher probability of being correctly recognized than a corresponding word in the second time-indexed text string.
8 more elements
A system for creating a hybrid text string, the system comprising:
- an initialization component configured to:
- receive a first time-indexed text string generated from an audio input by a first automated speech recognition system;
- and
- receive a second time-indexed text string generated from the audio input by a second automated speech recognition system;
- an anchor identification component configured to:
- identify an anchor word, wherein the anchor word is associated with an anchor word point, and wherein the anchor word point is located at a time location ti in the first time-indexed text string and at a time location t2 in the second time-indexed text string,
- wherein time locations t1 and t2 are different, and,
- wherein the anchor word is a word having a high probability of being correctly recognized by both the first and second automated speech recognition systems;
- a match vector calculation component configured to:
- calculate a match vector, wherein the match vector is associated with a variance between the time location t1 in the first time-indexed text string and the time location t2 in the second time-indexed text string;
- a word matching component configured to:
- identify a matched word, wherein the matched word is located at a time location t3 in the first time-indexed text string and at a time location t4 in the second time-indexed text string, wherein the time location t4 is located within a second time range in the second time-indexed text string, wherein the second time range is generated by a mathematical operation using at least the time location t3, the match vector, and a margin of error;
- and
- a string generation component configured to:
- create a hybrid text string, wherein the hybrid text string includes at least the anchor word and the matched word.
12 more elements
A method of creating a hybrid text string, wherein the method is performed by a computing system having a processor and a memory, the method comprising:
- receiving at least two different time-indexed text strings,
- wherein the two different time-indexed text strings are generated from the same audio input,
- wherein the audio input includes spoken words, and
- wherein two different time-indexed text strings were produced from either two different automated speech recognition systems, or the same automated speech recognition system with two different configurations or settings;
- correlating the two different time-indexed text strings based at least in part on a comparison of the two different time-indexed text strings to compensate for time differences between the two different time-indexed text strings;
- and,
- creating a hybrid text string, wherein the hybrid text string includes words that match between the two different time-indexed text strings, and
- when a word does not match between the two different time-indexed text strings, the hybrid text string includes a higher probability word from one of the two different time-indexed text strings.
5 more elements
A system for creating a text string, the system comprising:
- an initialization component configured to:
- receive a first time-indexed text string generated from an audio input by a first automated speech recognition system;
- receive a second time-indexed text string generated from the audio input by a second automated speech recognition system;
- and
- correlate the first and second time-indexed text strings based at least in part on a comparison between the first and second time-indexed text strings to compensate for time difference between the first and second time-indexed text strings;
- and
- a string generation component configured to:
- create an improved text string, wherein the improved text string is generated from a comparison between the first and second time-indexed text strings, and
- wherein the improved text string includes at least one word that matches between the first time-indexed text string and the second time-indexed text string.
6 more elements
A system for creating a hybrid text string, the system comprising:
- means for receiving first and second time-indexed text strings generated from an audio input by first and second automated speech recognition systems, respectively;
- means for identifying at least one anchor word, wherein the at least one anchor word is associated with an anchor word point,
- wherein the at least one anchor word is a word having a high probability of being correctly recognized by both the first and second automated speech recognition systems, and,
- wherein the anchor word point is located at a time location t1 in the first time-indexed text string and at a time location t2 in the second time-indexed text string, wherein the time locations tl and t2 are different;
- means for calculating a match vector, wherein the match vector is associated with a variance between the time location t1 in the first time-indexed text string and the time location t2 in the second time-indexed text string;
- means for identifying at least one matched word, wherein the at least one matched word is located at a time location t3 in the first time-indexed text string and at a differing time location t4 in the second time-indexed text string;
- and,
- means for creating a hybrid text string, wherein the hybrid text string includes the at least one anchor word and the at least one matched word, and wherein the hybrid text string includes at least one word from the first time-indexed text string that has a higher probability of being correctly recognized than a corresponding word in the second time-indexed text string.
5 more elements
Dependencies are inferred from the claim text.