Method and system for improving robustness in a speech system
View Patent ↗The present invention introduces a system and method for improved robustness in a speech system. In one embodiment, a method comprises receiving an utterance from an intended talker at a speech recognition system; computing a speaker verification score with a voice characteristic model associated and with the utterance; computing a speech recognition score associated with the utterance; and selecting a best hypothesis associated with the utterance and based on both the speaker verification score and the speech recognition score.
1. A method comprising:
receiving a first utterance from a speaker at an integrated speech and speaker recognition system;
generating a voice characteristic model for the speaker;
receiving a second utterance from the speaker at the speaker recognition system;
processing a portion of speech associated with the second utterance, wherein processing comprises,
computing a speaker verification score based on the voice characteristic model associated with the portion of speech,
computing a speech recognition score associated with the portion of speech, and
generating a combined score by combining the speaker verification score and the speech recognition score; and
selecting a best hypothesis from a plurality of hypotheses representing automatic speech recognition results of the second utterance,
based upon the combined score.
2. The method of claim 1 , wherein the portion of speech includes a word, a sentence, a syllable, or a frame.
3. The method of claim 1 , wherein said processing further comprises altering a search path in a Viterbi search used by a speech recognizer.
4. The method of claim 1 , further comprising using hotword speech recognition to identify the speaker.
5. The method of claim 1 , wherein the voice characteristic model includes a voice print, a personal profile and linguistic characteristics.
6. A system comprising:
a speech system; and
a speech input device connected to the speech system; wherein the speech system comprises,
a voice server, wherein the server includes an integrated speech and speaker recognizer that,
receives a first utterance from a speaker via the speech input device;
creates a voice characteristic model for the speaker;
receives a second utterance from the speaker via the speech input device;
processes a portion of speech associated with the second utterance, wherein the processor
computes a speaker verification score based on the voice characteristic model associated with the portion of speech,
computes a speech recognition score associated with the portion of speech, and
generates a combined score by combining the speaker verification score and the speech recognition score; and
selects a best hypothesis from a plurality of hypotheses representing automatic speech recognition results of the second utterance, based upon the combined score.
7. The system of claim 6 , wherein the speech input device comprises a cellular telephone, an analog telephone, a digital telephone, and a voice over internet protocol device.
8. The system of claim 6 , wherein the portion of speech includes a word, a sentence, a syllable, or a frame.
9. The system of claim 6 , wherein the server is further configured to alter a search path in a Viterbi search used by a speech recognizer.
10. An integrated speech and speaker recognition system comprising:
means for receiving a first utterance from a speaker;
means for generating a voice characteristic model for the speaker;
means for receiving a second utterance from the speaker at the speaker recognition system;
means for processing a portion of speech associated with the second utterance, wherein said means for processing comprises,
means for computing a speaker verification score based on the voice characteristic model associated with the portion of speech,
means for computing a speech recognition score associated with the portion of speech, and
means for generating a combined score by combining the speaker verification score and the speech recognition score; and
means for selecting a best hypothesis from a plurality of hypotheses representing automatic speech recognition results of the second utterance, based upon the combined score.
11. The system of claim 10 , wherein the portion of speech includes a word, a sentence, a syllable, or a frame.
12. The system of claim 10 , wherein the means for processing further comprises means for altering a search path in a Viterbi search used by a speech recognizer on the second utterance.
13. The system of claim 10 , further comprising means for using hotword speech recognition to identify the speaker.
14. The system of claim 10 , wherein the voice characteristic model includes a voice print, a personal profile and linguistic characteristics.
15. A machine-readable medium having stored thereon a plurality of instructions which, when executed by a machine, cause said machine to perform a process comprising:
receiving a first utterance from a speaker at an integrated speech and speaker recognition system;
generating a voice characteristic model for the speaker;
receiving a second utterance from the speaker at the speaker recognition system;
processing a portion of speech associated with the second utterance, wherein processing comprises,
computing a speaker verification score based on the voice characteristic model associated with the portion of speech,
computing a speech recognition score associated with the portion of speech, and
generating a combined score by combining the speaker verification score and the speech recognition score; and
selecting a best hypothesis from a plurality of hypotheses representing automatic speech recognition results of the second utterance, based upon the combined score.
16. The machine-readable medium of claim 15 wherein the portion of speech includes a word, a sentence, a syllable, or a frame.
17. The machine-readable medium of claim 15 , having stored thereon additional instructions when processing a portion of speech, said additional instructions when executed by a machine, cause said machine to perform altering a search path in a Viterbi search used by a speech recognizer.
18. The machine-readable medium of claim 15 , having stored thereon additional instructions which, when executed by the machine while identifying a speaker, cause said machine to use hotword speech recognition to identify the speaker.
19. The machine-readable medium of claim 15 , wherein the voice characteristic model includes a voice print, a personal profile and linguistic characteristics.
20. A method comprising:
receiving an utterance from a speaker at a speech recognition system;
computing a speaker verification score based on a voice characteristic model associated and with the utterance;
computing a speech recognition score associated with the utterance; and
selecting a best hypothesis from a plurality of hypotheses representing automatic speech recognition results of the utterance, based on both the speaker verification score and the speech recognition score.
21. The method of claim 20 , wherein the voice characteristic model is obtained from a voice model database.
22. The method of claim 20 , wherein the voice characteristic model is obtained from a first portion of the utterance.
23. A speech recognition system comprising:
a speaker verifier;
a speech recognizer connected to the speaker verifier; and
an input device connected to the speaker verifier and speech recognizer,
wherein the input device receives an utterance from a speaker; and
wherein the speech recognizer generates a recognition score associated with the utterance and generates a plurality of hypotheses representing automatic speech recognition results of the utterance, the speaker verifier generates a speaker verification score associated with the utterance; and the recognition score is combined with the verification score to select a best hypothesis of the plurality of hypotheses.
24. The speech recognition system of claim 23 , wherein the speech recognizer and speaker verifier are software entities residing on a speech server, and wherein the speech server comprises a processor, a bus connected to the processor, and memory connected to the bus that stores the software entities.
25. The speech recognition system of claim 24 , further comprising a database connected to the speech server, wherein the database stores a voice characteristic model of the speaker.