System and method for increasing recognition rates of in-vocabulary words by improving pronunciation modeling
View Patent ↗The present disclosure relates to systems, methods, and computer-readable media for generating a lexicon for use with speech recognition. The method includes receiving symbolic input as labeled speech data, overgenerating potential pronunciations based on the symbolic input, identifying potential pronunciations in a speech recognition context, and storing the identified potential pronunciations in a lexicon. Overgenerating potential pronunciations can include establishing a set of conversion rules for short sequences of letters, converting portions of the symbolic input into a number of possible lexical pronunciation variants based on the set of conversion rules, modeling the possible lexical pronunciation variants in one of a weighted network and a list of phoneme lists, and iteratively retraining the set of conversion rules based on improved pronunciations. Symbolic input can include multiple examples of a same spoken word. Speech data can be labeled explicitly or implicitly and can include words as text and recorded audio.
1. A method comprising:
receiving, by a processor, symbolic input as labeled speech data;
overgenerating potential pronunciations based on the symbolic input, wherein the overgenerating comprises:
establishing a set of conversion rules; and
converting portions of the symbolic input into a number of possible lexical pronunciation variants based on the set of conversion rules;
identifying potential pronunciations in a speech recognition context to yield identified potential pronunciations; and
storing the identified potential pronunciations in a lexicon.
2. The method of claim 1 , wherein symbolic input includes multiple examples of a same spoken word.
3. The method of claim 1 , wherein speech data is labeled explicitly.
4. The method of claim 1 , wherein speech data is labeled implicitly.
5. The method of claim 1 , wherein labeled speech data includes words as text and recorded audio.
6. The method of claim 1 , wherein overgenerating potential pronunciations further comprises:
modeling the possible lexical pronunciation variants in one of a weighted network and a list of phoneme lists.
7. The method of claim 1 , the method further comprising iteratively retraining the set of conversion rules based on improved pronunciations.
8. The method of claim 1 , wherein identifying potential pronunciations is based on a threshold.
9. A system for generating a lexicon for use with speech recognition, the system comprising:
a processor;
a first module configured to cause the processor to receive symbolic input as labeled speech data;
a second module configured to cause the processor to overgenerate potential pronunciations based on the symbolic input, wherein the second module further causes the processor to:
establish a set of conversion rules; and
convert portions of the symbolic input into a number of possible lexical pronunciation variants based on the set of conversion rules;
a third module configured to cause the processor to identify potential pronunciations in a speech recognition context to yield identified potential pronunciations; and
a fourth module configured to cause the processor to store the identified potential pronunciations in a lexicon.
10. The system of claim 9 , wherein symbolic input includes multiple examples of a same spoken word.
11. The system of claim 9 , wherein speech data is labeled explicitly.
12. The system of claim 9 , wherein speech data is labeled implicitly.
13. The system of claim 9 , wherein labeled speech data includes words as text and recorded audio.
14. The system of claim 9 , wherein the second module is further configured to control the processor to
model the possible lexical pronunciation variants in one of a weighted network and a list of phoneme lists.
15. The system of claim 9 , the system further comprising a fifth module configured to control the processor to iteratively retrain the set of conversion rules based on improved pronunciations.
16. A non-transitory computer-readable medium storing a computer program having instructions for generating a lexicon for use with speech recognition, the instructions comprising:
receiving symbolic input as labeled speech data;
overgenerating potential pronunciations based on the symbolic input, wherein the overgenerating comprises:
establishing a set of conversion rules; and
converting portions of the symbolic input into a number of possible lexical pronunciation variants based on the set of conversion rules;
identifying potential pronunciations in a speech recognition context to yield identified potential pronunciations; and
storing the identified potential pronunciations in a lexicon.
17. The non-transitory computer-readable medium of claim 16 , wherein symbolic input includes multiple examples of a same spoken word.
18. The non-transitory computer-readable medium of claim 16 , wherein speech data is labeled explicitly.
19. The non-transitory computer-readable medium of claim 16 , wherein speech data is labeled implicitly.
20. The non-transitory computer-readable medium of claim 16 , wherein overgenerating potential pronunciations further comprises:
modeling the possible lexical pronunciation variants in one of a weighted network and a list of phoneme lists.