Generating topic-specific language models
Speech recognition may be improved by generating and using a topic specific language model. A topic specific language model may be created by performing an initial pass on an audio signal using a generic or basis language model. A speech recognition device may then determine topics relating to the audio signal based on the words identified in the initial pass and retrieve a corpus of text relating to those topics. Using the retrieved corpus of text, the speech recognition device may create a topic specific language model. In one example, the speech recognition device may adapt or otherwise modify the generic language model based on the retrieved corpus of text.
1. A method comprising:
determining, based on a speech recognition process associated with a first language model, a) a first probability value of a second word appearing directly following a first word in a phrase or sentence, and b) a topic associated with an audio signal;
performing a plurality of searches of a corpus to identify a plurality of terms related to the topic, wherein the corpus comprises a collection of text other than a transcript of the audio signal;
in response to determining that a quantity of the plurality of terms identified by the searches as related to the topic matches or exceeds a threshold quantity: generating, based on the plurality of terms identified in the corpus, a second language model;
determining, based on the second language model, a second probability value of the second word appearing directly following the first word; and
in response to determining that the second probability value is not the same as the first probability value: modifying the first language model to reflect the second probability value.
2. The method of claim 1 , further comprising:
determining an average value based on an average of the first and the second probability value; and
using the average probability value to determine probability of the second word appearing directly following the first word.
3. The method of claim 1 , further comprising, using the modified first language model that reflects the second probability value to determine probability of a sequence of words appearing together.
4. The method of claim 1 , further comprising, using the modified first language model in automatically transcribing a recorded audio.
5. The method of claim 1 , further comprising, further modifying the first language model based on frequency of occurrence of word combinations in a collected corpus of text.
6. The method of claim 5 , further comprising, using the further modified first language model based on the frequency of occurrence of word combinations as a second pass on the speech recognition process to identify words used in a speech.
7. The method of claim 1 , wherein the plurality of searches of the corpus comprise searching on the internet.
8. The method of claim 1 , wherein the plurality of searches of the corpus comprises searching a publication database.
9. The method of claim 1 , wherein the quantity of the plurality of terms is based on at least one of:
a total quantity of terms needed to generate the second language model and a quantity of topics associated with the audio signal; or
a respective significance, based on the speech recognition process, for each of a plurality of topics associated with the audio signal.
10. The method of claim 1 , further comprising, in response to determining that the quantity of the plurality of terms identified by the searches as related to the topic does not match or exceeds a threshold quantity: continuing to perform searches to identify terms until corresponding search results matches or exceeds the threshold quantity.
11. An apparatus comprising:
one or more processors; and
memory storing instructions that, when executed by the one or more processors, cause the apparatus to:
determine, based on a speech recognition process associated with a first language model, a) a first probability value of a second word appearing directly following a first word in a phrase or sentence, and b) a topic associated with an audio signal;
perform a plurality of searches of a corpus to identify a plurality of terms related to the topic, wherein the corpus comprises a collection of text other than a transcript of the audio signal;
in response to determining that a quantity of the plurality of terms identified by the searches as related to the topic matches or exceeds a threshold quantity: generate, based on the plurality of terms identified in the corpus, a second language model;
determine, based on the second language model, a second probability value of the second word appearing directly following the first word; and
in response to determining that the second probability value is not the same as the first probability value: modify the first language model to reflect the second probability value.
12. The apparatus of claim 11 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to:
determine an average value based on an average of the first and the second probability value; and
use the average probability value to determine probability of the second word appearing directly following the first word.
13. The apparatus of claim 11 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to use the modified first language model that reflects the second probability value to determine probability of a sequence of words appearing together.
14. The apparatus of claim 11 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to use the modified first language model in automatically transcribing a recorded audio.
15. The apparatus of claim 11 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to further modify the first language model based on frequency of occurrence of word combinations in a collected corpus of text.
16. The apparatus of claim 15 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to use the further modified first language model based on the frequency of occurrence of word combinations as a second pass on the speech recognition process to identify words used in a speech.
17. The apparatus of claim 11 , wherein the plurality of searches of the corpus comprise searching on the internet.
18. The apparatus of claim 11 , wherein the plurality of searches of the corpus comprises searching a publication database.
19. The apparatus of claim 11 , wherein the quantity of the plurality of terms is based on at least one of:
a total quantity of terms needed to generate the second language model and a quantity of topics associated with the audio signal; or
a respective significance, based on the speech recognition process, for each of a plurality of topics associated with the audio signal.
20. The apparatus of claim 11 , wherein the instructions, when executed by the one or more processors, further cause the apparatus to, in response to determining that the quantity of the plurality of terms identified by the searches as related to the topic does not match or exceeds a threshold quantity: continue performing searches to identify terms until corresponding search results matches or exceeds the threshold quantity.