Automated word correction in speech recognition systems
Systems and methods for correcting recognition errors in speech recognition systems are disclosed herein. Natural conversational variations are identified to determine whether a query intends to correct a speech recognition error or whether the query is a new command. When the query intends to correct a speech recognition error, the system identifies a location of the error and performs the correction. The corrected query can be presented to the user or be acted upon as a command for the system.
1. A method for correcting a speech recognition error, the method comprising:
detecting a first voice input;
generating a text string based on the first voice input;
detecting a second voice input, wherein the second voice input was input within a predetermined time threshold of the generating the text string based on the first voice input;
determining whether the second voice input is directed to correcting the text string generated based on the first voice input, by:
identifying a first plurality of sound properties corresponding to the first voice input;
identifying a second plurality of sound properties corresponding to the second voice input;
comparing each of the first plurality of sound properties of the first voice input to each corresponding one of the second plurality of sound properties of the second voice input; and
determining, based on the comparing, whether the second voice input is directed to correcting the text string previously generated based on the first voice input when a number of the first plurality of sound properties that matches the second plurality of sound properties is greater than a predetermined threshold and at least one of the first plurality of sound properties does not match the corresponding one of the second plurality of sound properties; and
in response to the determining that the second voice input is directed to correcting the text string, modifying the text string based on the second voice input.
2. The method of claim 1 , wherein determining whether the second voice input is directed to correcting the text string comprises:
determining that at least one of an acoustic envelope, an intensity, a pitch, a frequency, or an amplitude of the second voice input exceeds a threshold.
3. The method of claim 1 , wherein determining that at least the portion of the first plurality of sound properties matches the at least the portion of the second plurality of sound properties comprises determining that an acoustic envelope of the at least the portion of the second plurality of sound properties matches an acoustic envelope of the at least the portion of the first plurality of sound properties.
4. The method of claim 1 , wherein the text string previously generated based on the first voice input is a first text string, and wherein determining whether the second voice input is directed to correcting the first text string comprises:
converting the second voice input to a second text string; and
determining, based on a database of misrecognized words, that a first word in the first text string and a second word in the second text string are misrecognized for one another.
5. The method of claim 4 , wherein modifying the first text string based on the second voice input comprises replacing the first word in the first text string with the second word.
6. The method of claim 1 , further comprising:
identifying, based at least in part on the sound property of the second voice input, a portion of the text string to be modified,
wherein the modifying the text string comprises modifying only the identified portion of the text string.
7. The method of claim 1 , wherein the sound property of the voice input comprises at least one of an acoustic envelope, an intensity, a pitch, a frequency, or an amplitude.
8. The method of claim 1 , wherein the sound property corresponds to a correction expression.
9. A system for correcting a speech recognition error, the system comprising:
memory; and
control circuitry configured to:
detect a first voice input;
generate a text string based on the first voice input;
detect a second voice input, wherein the second voice input was input within a predetermined time threshold of the generating the text string based on the first voice input;
determine whether the second voice input is directed to correcting the text string previously generated based on the first voice input, by:
identifying a first plurality of sound properties corresponding to the first voice input;
identifying a second plurality of sound properties corresponding to the second voice input;
compare each of the first plurality of sound properties of the first voice input to each corresponding one of the second plurality of sound properties of the second voice input; and
determining, based on the comparing, whether the second voice input is directed to correcting the text string previously generated based on the first voice input when a number of the first plurality of sound properties that matches the second plurality of sound properties is greater than a predetermined threshold and at least one of the first plurality of sound properties does not match the corresponding one of the second plurality of sound properties; and
in response to the determining that the second voice input is directed to correcting the text string, modify the text string based on the second voice input.
10. The system of claim 9 , wherein the control circuitry is configured to determine whether the second voice input is directed to correcting the text string by:
determining that at least one of an acoustic envelope, an intensity, a pitch, a frequency, or an amplitude of the second voice input exceeds a threshold.
11. The system of claim 9 , wherein the control circuitry is configured to determine that at least the portion of the first plurality of sound properties matches the at least the portion of the second plurality of sound properties by determining that an acoustic envelope of the at least the portion of the second plurality of sound properties matches an acoustic envelope of the at least the portion of the first plurality of sound properties.
12. The system of claim 9 , wherein the text string previously generated based on the first voice input is a first text string, and wherein the control circuitry is configured to determine whether the second voice input is directed to correcting the first text string by:
converting the second voice input to a second text string; and
determining, based on a database of misrecognized words, that a first word in the first text string and a second word in the second text string are misrecognized for one another.
13. The system of claim 12 , wherein the control circuitry is configured to modify the first text string based on the second voice input by replacing the first word in the first text string with the second word.
14. The system of claim 9 , wherein the control circuitry is further configured to:
identify, based at least in part on the sound property of the second voice input, a portion of the text string to be modified,
wherein the modifying the text string comprises modifying only the identified portion of the text string.
15. The system of claim 9 , wherein the sound property of the voice input comprises at least one of an acoustic envelope, an intensity, a pitch, a frequency, or an amplitude.
16. The system of claim 9 , wherein the sound property corresponds to a correction expression.
17. A method for correcting a speech recognition error, the method comprising:
detecting a first voice input;
generating a text string based on the first voice input;
detecting a second voice input, wherein the second voice input was input within a predetermined time threshold of the generating the text string based on the first voice input;
identifying a first plurality of sound properties corresponding to the first voice input;
identifying a second plurality of sound properties corresponding to the second voice input;
comparing each of the first plurality of sound properties of the first voice input to each corresponding one of the second plurality of sound properties of the second voice input;
determining, based on the comparing, whether a number of the first plurality of sound properties that matches the second plurality of sound properties is greater than a predetermined threshold and at least one of the first plurality of sound properties does not match the corresponding one of the second plurality of sound properties; and
in response to the determining that the number of the first plurality of sound properties matches the second plurality of sound properties is greater than the predetermined threshold and at least one of the first plurality of sound properties does not match the corresponding one of the second plurality of sound properties, modifying the text string based on the second voice input.