Adaptive multi-pass speech recognition system

• US 7,401,017 B2
• Filed: 04/04/2006
• Issued: 07/15/2008
• Est. Priority Date: 04/20/1999
• Status: Expired due to Fees

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Kannan, Ashvin, Murveit, Hy, Leggetter, Chris, Shahshahani, Ben, Knill, Katherine
• Primary Examiner(s): Opsasnick; Michael N.
• Application Number:
  US11/398,327
• Publication Number:
  US 20060178879A1
• Time in Patent Office: 833 Days
• Field of Search: 704246-250
• US Class Current: 704/247
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 2015/085 Methods for reducing search complexity, pruning

Abstract

Method and apparatus for multi-pass speech recognition. An input device receives spoken input. A processor performs a first pass speech recognition technique on the spoken input and forms first pass results. The first pass results include a number of alternative speech expressions, each having an assigned score related to the certainty that the corresponding expression correctly matches the spoken input. The processor selectively performs a second pass speech recognition technique on the spoken input according to the first pass results. Preferably, the second pass attempts to correctly match the spoken input to only those expressions which were identified during the first pass. Otherwise, if one of the expressions identified by the first pass is assigned a score higher than a predetermined threshold (e.g., 95%), the second pass is not performed. Because the second pass is performed only when necessary, the invention recognizes speech with a faster average speed for a given accuracy in comparison to prior systems. Alternately, the first pass results identify a characteristic of the spoken input. The characteristic can be the gender of the speaker or a type of telephone the speaker is calling from. In which case, the second pass speech recognition technique is selected from a plurality of speech recognition techniques according to the characteristic identified by the first pass. Because the selected second pass technique is specific to the characteristic of the spoken input, the second pass technique can perform speech recognition faster for a given accuracy than a technique which is not specific.

Claims

1. A speech recognition system for recognizing spoken input received from a source of the spoken input coupled to the speech recognition system wherein the speech recognition system comprises:

a. input means for receiving the spoken input from the source of the spoken input; and

b. processing means coupled to the input means for performing a first pass speech recognition technique on the spoken input and for forming first pass results;

wherein the first pass results identify a speech expression as corresponding to the spoken input with a corresponding score and wherein a characteristic of the spoken input is identified based upon the first pass results, the processing means selectively performs a second pass speech recognition technique on the spoken input according to the first pass results, wherein the second pass speech recognition technique is performed on the spoken input when the corresponding score is below a predetermined threshold and the second pass speech recognition technique is selected from a plurality of speech recognition techniques according to the characteristic of the spoken input identified by the first pass, and further the characteristic of the spoken input is gender of a speaker of the spoken input.

2. A speech recognition system for recognizing spoken input received from a source of the spoken input coupled to the speech recognition system wherein the speech recognition system comprises:

a. input means for receiving the spoken input from the source of the spoken input; and

b. processing means coupled to the input means for performing a first pass speech recognition technique on the spoken input and for forming first pass results;

wherein the first pass results identify a speech expression as corresponding to the spoken input with a corresponding score and wherein the processing means selectively performs a second pass speech recognition technique on the spoken input according to the first pass results, wherein the second pass speech recognition technique is performed on the spoken input when the corresponding score is below a predetermined threshold and further the first pass results identify the spoken input as being in one of the following three categories;

(1) originating from a female speaker;

(2) originating from a male speaker; and

(3) originating from a hands-free telephone where a speaker of the spoken input is female or male.

3. A method of recognizing spoken input received from a source of the spoken input wherein the method comprises steps of:

a. receiving the spoken input from the source of the spoken input;

b. performing a first pass speech recognition technique on the spoken input;

c. forming first pass results wherein the first pass results identify a speech expression as corresponding to the spoken input with a corresponding score; and

d. selectively performing a second pass speech recognition technique on the spoken input according to the first pass results;

wherein the second pass speech recognition technique is performed on the spoken input when the corresponding score is below a predetermined threshold and wherein the first pass results identify a characteristic of the spoken input, the second pass speech recognition technique is selected from a plurality of speech recognition techniques according to the characteristic of the spoken input identified by the first pass, and further the characteristic of the spoken input is gender of a speaker of the spoken input.

4. A method of recognizing spoken input received from a source of the spoken input wherein the method comprises steps of:

a. receiving the spoken input from the source of the spoken input;

b. performing a first pass speech recognition technique on the spoken input;

c. forming first pass results wherein the first pass results identify a speech expression as corresponding to the spoken input with a corresponding score; and

d. selectively performing a second pass speech recognition technique on the spoken input according to the first pass results;

wherein the second pass speech recognition technique is performed on the spoken input when the corresponding score is below a predetermined threshold and wherein the first pass results identify the spoken input as being in one of the following three categories;

(1) originating from a female speaker;

(2) originating from a male speaker; and

(3) originating from a hands-free telephone where a speaker of the spoken input is female or male.

5. A method of recognizing spoken input received from a source of the spoken input wherein the method comprises steps of:

a. receiving the spoken input from the source of the spoken input;

b. performing a first pass speech recognition technique on the spoken input;

c. forming first pass results wherein the first pass results identify a speech expression as corresponding to the spoken input with a corresponding score and also identify a characteristic of the spoken input; and

d. performing a second pass speech recognition technique on the spoken input when the corresponding score is below a predetermined threshold and wherein the second pass speech recognition technique is selected from a plurality of speech recognition techniques according to the identified characteristic;

wherein the characteristic of the spoken input is gender of a speaker of the spoken input.

6. A method of recognizing spoken input received from a source of the spoken input wherein the method comprises steps of:

a. receiving the spoken input from the source of the spoken input;

b. performing a first pass speech recognition technique on the spoken input;

c. forming first pass results wherein the first pass results identify a speech expression as corresponding to the spoken input with a corresponding score and also identify a characteristic of the spoken input; and

d. performing a second pass speech recognition technique on the spoken input when the corresponding score is below a predetermined threshold and wherein the second pass speech recognition technique is selected from a plurality of speech recognition techniques according to the identified characteristic;

wherein the characteristic of the spoken input is a type of telephone channel a speaker of the spoken input is calling from.

7. A method of recognizing spoken input received from a source of the spoken input wherein the method comprises steps of:

a. receiving the spoken input from the source of the spoken input;

b. performing a first pass speech recognition technique on the spoken input;

c. forming first pass results wherein the first pass results identify a plurality of speech expressions as corresponding to the spoken input each speech expression having a corresponding score and also identify a characteristic of the spoken input; and

d. performing a second pass speech recognition technique on the spoken input when a difference between two highest of the scores does not exceed a predetermined threshold and wherein the second pass speech recognition technique is selected from a plurality of speech recognition techniques according to the identified characteristic;

wherein the characteristic of the spoken input is gender of a speaker of the spoken input.

8. A method of recognizing spoken input received from a source of the spoken input wherein the method comprises steps of:

a. receiving the spoken input from the source of the spoken input;

b. performing a first pass speech recognition technique on the spoken input;

c. forming first pass results wherein the first pass results identify a plurality of speech expressions as corresponding to the spoken input each speech expression having a corresponding score and also identify a characteristic of the spoken input; and

d. performing a second pass speech recognition technique on the spoken input when a difference between two highest of the scores does not exceed a predetermined threshold and wherein the second pass speech recognition technique is selected from a plurality of speech recognition techniques according to the identified characteristic;

wherein the characteristic of the spoken input is a type of telephone channel a speaker of the spoken input is calling from.