System and method for tightly coupling automatic speech recognition and search
Systems, methods, and computer-readable storage media relate to performing a search. A system configured to practice the method first receives from an automatic speech recognition (ASR) system a word lattice based on speech query and receives indexed documents from an information repository. The system composes, based on the word lattice and the indexed documents, at least one triple including a query word, selected indexed document, and weight. The system generates an N-best path through the word lattice based on the at least one triple and re-ranks ASR output based on the N-best path. The system aggregates each weight across the query words to generate N-best listings and returns search results to the speech query based on the re-ranked ASR output and the N-best listings. The lattice can be a confusion network, the arc density of which can be adjusted for a desired performance level.
1. A method comprising:
receiving, from an automatic speech recognition system, a word lattice based on a speech query;
composing a triple comprising a query word from the speech query, an indexed document, and a weight;
generating an N-best path through the word lattice based on the triple;
re-ranking automatic speech recognition output based on the N-best path, to yield re-ranked automatic speech recognition output; and
returning search results to the speech query based on the re-ranked automatic speech recognition output.
2. The method of claim 1 , further comprising:
aggregating each weight across words in the speech query to generate N-best listings; and
wherein the search results are further based on the N-best listings.
3. The method of claim 1 , wherein the word lattice is a word confusion network.
4. The method of claim 3 , further comprising adjusting an arc density of the word confusion network based on a desired performance level.
5. The method of claim 3 , further comprising adjusting an arc density of the word confusion network based on a desired accuracy level.
6. The method of claim 2 , wherein the weight is determined based on a plurality of relevance metrics.
7. The method of claim 1 , wherein a mobile device receives the speech query.
8. The method of claim 1 , wherein re-ranking of the automatic speech recognition output is based on constraints encoded by relevance metrics and the information repository.
9. The method of claim 1 , wherein the indexed document is represented as a search finite state machine.
10. The method of claim 9 , wherein the search finite state machine represents an index of the indexed document.
11. A system comprising:
a processor; and
a computer-readable storage medium having instructions stored which, when executed by the processor, cause the processor to perform operations comprising:
receiving, from an automatic speech recognition system, a word lattice based on a speech query;
composing a triple comprising a query word from the speech query, an indexed document, and a weight;
generating an N-best path through the word lattice based on the triple;
re-ranking automatic speech recognition output based on the N-best path, to yield re-ranked automatic speech recognition output; and
returning search results to the speech query based on the re-ranked automatic speech recognition output.
12. The system of claim 11 , the computer-readable storage medium having instructions stored which, when executed by the processor, result in operations comprising:
aggregating each weight across words in the speech query to generate N-best listings; and
wherein the search results are further based on the N-best listings.
13. The system of claim 11 , wherein the word lattice is a word confusion network.
14. The system of claim 13 , the computer-readable storage medium having instructions stored which, when executed by the processor, result in operations comprising adjusting an arc density of the word confusion network based on a desired performance level.
15. The system of claim 13 , the computer-readable storage medium having instructions stored which, when executed by the processor, result in operations comprising adjusting an arc density of the word confusion network based on a desired accuracy level.
16. The system of claim 11 , wherein the weight is determined based on a plurality of relevance metrics.
17. The system of claim 11 , wherein a mobile device receives the speech query.
18. The system of claim 11 , wherein re-ranking of the automatic speech recognition output is based on constraints encoded by relevance metrics and the information repository.
19. The system of claim 11 , wherein the indexed document is represented as a search finite state machine.
20. A computer-readable storage device having instructions stored which, when executed by a computing device, cause the computing device to perform operations comprising:
receiving, from an automatic speech recognition system, a word lattice based on a speech query;
composing a triple comprising a query word from the speech query, an indexed document, and a weight;
generating an N-best path through the word lattice based on the triple;
re-ranking automatic speech recognition output based on the N-best path, to yield re-ranked automatic speech recognition output; and
returning search results to the speech query based on the re-ranked automatic speech recognition output.