Speech recognition models combining gender-dependent and gender-independent phone states and using phonetic-context-dependence

• US 5,953,701 A
• Filed: 01/22/1998
• Issued: 09/14/1999
• Est. Priority Date: 01/22/1998
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Roukos, Salim Estephan, Neti, Chalapathy Venkata
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Smits, Talivaldis Ivars
• Application Number:
  US09/010,466
• Time in Patent Office: 600 Days
• Field of Search: 704/240, 704/242, 704/252, 704/254
• US Class Current: 704/254
• CPC Class Codes: G10L 15/07 to the speaker
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]

Abstract

A method of gender dependent speech recognition includes the steps of identifying phone state models common to both genders, identifying gender specific phone state models, identifying a gender of a speaker and recognizing acoustic data from the speaker. A method of constructing a gender-dependent speech recognition model includes the steps of providing training data of a known gender, aligning the training data, tagging the training data with a gender to create gender-tagged data, determining a gender question at a node to determine gender dependence of the gender-tagged data, determining a phonetic context question at the node to determine phonetic context dependence of the gender-tagged data, determining a highest value of an evaluation function between the gender dependence and the phonetic context dependence to determine which dependence is a dominant dependence, splitting the data of the dominant dependence into child nodes according to likelihood criteria, comparing the highest value with a threshold value to determine if additional splitting is necessary, repeating theses steps for each child node until the highest value is below the threshold value and counting the nodes having gender dependence to determine an overall gender dependence level. A gender-dependent speech recognition system includes an input device for inputting speech to a preprocessor. The preprocessor converts the speech into acoustic data, and a processor for identifies gender-dependent phone state models and phone state modes common to both genders. The phone state models are stored in a memory device wherein the processor recognizes the speech in accordance with the phone state models.

Claims

1. A method of gender dependent speech recognition comprising the steps of:

identifying phone state models common to both genders;

identifying gender specific phone state models;

identifying a gender of a speaker; and

recognizing acoustic data from the speaker based on the phone state models.

2. The method of gender dependent speech recognition as recited in claim 1, wherein the step of identifying gender further comprises the steps of:

setting the gender to a first gender;

calculating a confidence based on the first gender;

setting the gender to a second gender;

calculating a confidence of the second gender; and

determining gender by selecting the confidence with a higher value.

3. The method of gender dependent speech recognition as recited in claim 2, wherein a likelihood for each gender is used to determine gender.

4. The method of gender dependent speech recognition as recited in claim 2, wherein the steps of calculating the confidences include calculating the confidence by taking the product of word confidences for every word in a hypothesis.

5. The method of gender dependent speech recognition as recited in claim 1, wherein the step of identifying gender includes performing a maximum a posteriori adaptation.

6. The method of gender dependent speech recognition as recited in claim 1, wherein the step of identifying gender includes comparing Gaussian prototypes to a codebook of Gaussian prototypes to determine gender.

7. The method of gender dependent speech recognition as recited in claim 6, wherein the step of comparing Gaussian prototypes includes clustering Gaussian prototypes to create the codebook of Gaussian prototypes.

8. The method of gender dependent speech recognition as recited in claim 1, wherein the step of identifying gender specific phone state models further comprises the step of asking a gender question at a node to determine gender dependence of the acoustic data.

9. A method of constructing a gender-dependent speech recognition model comprising the steps of:

a) aligning acoustic data with a gender independent system

b) asking a gender question at a node to determine gender dependence of the acoustic data;

c) asking a phonetic context question at the node to determine phonetic context dependence of the acoustic data;

d) determining a highest value of an evaluation function between the gender dependence and the phonetic context dependence to determine which dependence is a dominant dependence;

e) splitting the data of the dominant dependence into child nodes according to the question of dominant dependence; and

f) repeating steps b-e for each child node until a threshold criterion is met.

10. The method of constructing a gender-dependent speech recognition model as recited in claim 9, further comprises the step of counting the nodes having gender dependence to determine an overall gender dependence level.

11. The method of constructing a gender-dependent speech recognition model as recited in claim 9, wherein the step of repeating steps until a threshold criterion is met includes:

comparing the highest value with a threshold value to determine if additional splitting is necessary.

12. The method of constructing a gender-dependent speech recognition model as recited in claim 9, wherein the step of asking the phonetic context question includes the step of asking the phonetic context question at each position between -5 to +5, inclusive from the node.

13. A method of constructing a gender-dependent speech recognition model comprising the steps of:

a) providing training data of a known gender;

b) aligning the training data;

c) tagging the training data with a gender to create gender-tagged data;

d) asking a gender question at a node to determine gender dependence of the gender-tagged data;

e) asking a phonetic context question at the node to determine phonetic context dependence of the gender-tagged data;

f) determining a highest value of an evaluation function between the gender dependence and the phonetic context dependence to determine which dependence is a dominant dependence;

g) splitting the data of the dominant dependence into child nodes according to a likelihood criterion;

h) comparing the highest value with a threshold value to determine if additional splitting is necessary; and

i) repeating steps d-h for each child node until the highest value is below the threshold value.

14. The method of constructing a gender-dependent speech recognition model as recited in claim 13, further comprises the step of counting the nodes having gender dependence to determine an overall gender dependence level.

15. The method of constructing a gender-dependent speech recognition model as recited in claim 13, wherein the step of asking the phonetic context question includes the step of asking the phonetic context question at each position between -5 to +5, inclusive from the node.

16. A gender-dependent speech recognition system comprising:

an input device for inputting speech to a preprocessor, the preprocessor converting speech into acoustic data; and

a processor for identifying gender-dependent phone state models and phone state modes common to both genders, the phone state models being stored in a memory device wherein the processor recognizes the speech in accordance with the phone state models.

17. The gender-dependent speech recognition system as recited in claim 16, wherein the gender-dependent phone state models reduce an amount of memory storage space needed to store the phone state models.

18. The gender-dependent speech recognition system as recited in claim 16, wherein the processor includes a computer.