Regional features based speech recognition method and system

1. A regional-features-based speech recognition method, comprising:

learning speech features by region using speech data classified by region category; and

recognizing, by a speech recognition device, input speech using an acoustic model and a language model generated through classification of a region category for the input speech and the learning,

wherein the recognizing the input speech comprises:

inputting the input speech into a plurality of by-region speech recognizers using the acoustic model and by-region language models, each of the plurality of by-region speech recognizers corresponding to a different regional dialect;

generating output results based on the input speech by each of the plurality of by-region speech recognizers;

generating a prediction of a region category corresponding to the input speech based on probability values assigned to the output results based on an accent of the input speech, a characteristic word by region, and a number of times that the characteristic word is used in a speech recognition process; and

selecting a speech recognition result output by a by-region speech recognizer among the plurality of by-region speech recognizers that has received a highest score according to the prediction,

wherein the learning speech features by region further comprises training the language model based on a corpus collected by region,

wherein the training the language model comprises performing region information vector labeling by word with respect to words included in the corpus collected by region,

wherein the training the language model comprises performing vector labeling in which, with respect to a word frequently used by region, a value of information of a corresponding region is set to be high, and

wherein the generating the prediction of the region category comprises ranking outputs of the plurality of by-region speech recognizers based on an accent of the input speech, a word included in the input speech and having regional characteristics, and a number of times that the word is used.

2. The regional-features-based speech recognition method according to claim 1 , wherein the learning speech features by region comprises classifying speech features by region category based on accent.

3. The regional-features-based speech recognition method according to claim 1 , wherein the learning speech features by region comprises generating a region classification learning model using extracted speech features.

4. The regional-features-based speech recognition method according to claim 2 , wherein the learning speech features by region further comprises training the acoustic model using the classified speech features.

5. The regional-features-based speech recognition method according to claim 1 , wherein the recognizing the input speech is performed in parallel by the plurality of by-region speech recognizers.

6. The regional-features-based speech recognition method according to claim 1 , further comprising:

ranking the plurality of by-region speech recognizers based on a sum of an accent-based region classification result probability vector, a by-word region information probability vector and a scalar value based on a recognition result word frequency.

7. The regional-features-based speech recognition method according to claim 1 , wherein the plurality of by-region speech recognizers include two or more of a British English speech recognizer, an American English speech recognizer, an Australian English speech recognizer and an Irish English speech recognizer.

8. A regional-features-based speech recognition system, comprising:

a learning processor configured to learn speech features by region using speech data classified by region category; and

a speech recognizer configured to recognize input speech using an acoustic model and a language model generated through classification of a region category for the input speech and the learning,

wherein the speech recognition system is further configured to:

input the input speech into a plurality of by-region speech recognizers using the acoustic model and by-region language models, each of the plurality of by-region speech recognizers corresponding to a different regional dialect,

generate output results based on the input speech by each of the plurality of by-region speech recognizers,

generate a prediction of a region category corresponding to the input speech based on probability values assigned to the output results based on an accent of the input speech, a characteristic word by region, and a number of times that the characteristic word is used in a speech recognition process, and

select a speech recognition result output by a by-region speech recognizer among the plurality of by-region speech recognizers that has received a highest score according to the prediction,

wherein the learning processor further comprises a language model learning processor configured to train the language model based on a corpus collected by region,

wherein the language model learning processor performs region information vector labeling by word with respect to words included in the corpus collected by region,

wherein the language model learning processor performs vector labeling in which, with respect to a word frequently used by region, a value of information of a corresponding region is set to be high, and

wherein the regional-features-based speech recognition system ranks outputs of by-region speech recognizers using a region classification result probability vector based on an accent of the input speech, a word included in the input speech and having regional characteristics, and a number of times that the word is used, a region information probability vector by word having regional characteristics, and a scalar value of the number of times.

9. The regional-features-based speech recognition system according to claim 8 , wherein the learning processor comprises a region classification learning processor configured to classify speech features by region category based on accent.

10. The regional-features-based speech recognition system according to claim 9 , wherein the region classification learning processor generates a region classifier configured to classify speech features extracted by region category based on accent.

11. The regional-features-based speech recognition system according to claim 9 , wherein the learning processor further comprises an acoustic model learning processor configured to train the acoustic model using the classified speech features.

12. The regional-features-based speech recognition system according to claim 8 , wherein the speech recognizer comprises by-region speech recognizers configured to recognize speech in parallel.