Enhancing signature word detection in voice assistants
Systems and methods detecting a spoken sentence in a speech recognition system are disclosed herein. Speech data is buffered based on an audio signal captured at a computing device operating in an active mode. The speech data is buffered irrespective of whether the speech data comprises a signature word. The buffered speech data is processed to detect a presence of the sentence comprising at least one command and a query for the computing device. Processing the buffered speech data includes detecting the signature word in the buffered speech data, and in response to detecting the signature word in the speech data, initiating detection of the sentence in the buffered speech data.
1. A method for detecting a sentence including at least one of a command and a query in a speech recognition system, the method comprising:
detecting, while a computing device is in a non-active mode, a user activity that suggests a user intention to interact with the computing device, wherein speech data is not buffered by the computing device in the non-active mode;
in response to detecting the user activity, enabling an active mode and buffering speech data based on an audio signal captured at the computing device operating in the active mode, wherein the speech data is buffered in the active mode irrespective of whether the speech data comprises a signature word; and
processing the buffered speech data to detect a presence of the sentence comprising at least one of the command and the query for the computing device, wherein processing the buffered speech data comprises:
detecting the signature word in the buffered speech data;
in response to detecting the signature word in the speech data, initiating detection of the sentence in the buffered speech data; and
detecting the sentence by identifying a beginning portion of the sentence in the buffered speech data and determining that the beginning portion of the sentence precedes a portion of the sentence corresponding to the signature word.
2. The method of claim 1 , further comprising detecting the sentence based on a sequence validating technique or based on a model trained to distinguish between user commands and user assertions.
3. The method of claim 1 , further comprising detecting the sentence by detecting silent durations occurring before and after, respectively, the sentence in the speech data, wherein detecting the silent durations is based on speech amplitude heuristics of the speech data.
4. The method of claim 1 , wherein detecting the signature word is performed at the computing device or at a server remote from the computing device.
5. The method of claim 1 , further comprising transmitting the speech data to a speech recognition processor for performing automated speech recognition (ASR) on the speech data.
6. The method of claim 1 , further comprising identifying the beginning portion of the sentence and an end portion of the sentence based on a trained model selected from one of a hidden Markov model (HMM), a long short-term memory (LSTM) model, and a bidirectional LSTM.
7. The method of claim 1 , wherein detecting the signature word is based on heuristics of audio signatures of a demographic region.
8. The method of claim 1 , wherein the computing device operates in the active mode only in response to receiving a user consent.
9. The method of claim 1 , wherein detecting the user activity comprises determining that a user has turned to face the computing device.
10. A system for detecting a sentence including at least one of a command and a query in a speech recognition system, the system comprising:
a memory; and
control circuitry communicatively coupled to the memory and configured to:
detect, while a computing device is in a non-active mode, a user activity that suggests a user intention to interact with the computing device, wherein speech data is not buffered by the computing device in the non-active mode;
in response to detecting the user activity, enable an active mode and buffer in the memory speech data based on an audio signal captured at the computing device operating in the active mode, wherein the speech data is buffered in the active mode irrespective of whether the speech data comprises a signature word; and
process the buffered speech data to detect a presence of the sentence comprising at least one of the command and the query for the computing device, wherein in processing the buffered speech data, the control circuitry is configured to:
detect the signature word in the buffered speech data;
in response to detecting the signature word in the speech data, initiate detection of the sentence in the buffered speech data; and
detect the sentence by identifying a beginning portion of the sentence in the buffered speech data and determining that the beginning portion of the sentence precedes a portion of the sentence corresponding to the signature word.
11. The system of claim 10 , wherein the control circuitry is further configured to detect the sentence based on a sequence validating technique or based on a model trained to distinguish between user commands and user assertions.
12. The system of claim 10 , wherein the control circuitry is further configured to:
detect the signature word by detecting silent durations occurring before and after, respectively, the sentence in the speech data; and the
detect the silent durations based on speech amplitude heuristics of the speech data.
13. The system of claim 10 , wherein the memory is local to the computing device.
14. The system of claim 10 , wherein the control circuitry is further configured to detect the signature word at the computing device.
15. The system of claim 10 , wherein the control circuitry is configured to transmit the buffered data packets to a remotely located server to detect the signature word.
16. The system of claim 10 , wherein the control circuitry is further configured to transmit the speech data to a speech recognition processor for performing automated speech recognition (ASR) on the speech data.
17. The system of claim 10 , wherein the control circuitry is further configured to identify the beginning portion of the sentence and an end portion of the sentence based on a trained model selected from one of a hidden Markov model (HMM), a long short-term memory (LSTM) model, and a bidirectional LSTM.
18. The system of claim 10 , wherein detection of the signature word is based on heuristics of audio signatures of a demographic region.
19. The system of claim 10 , wherein the control circuitry is configured to operate in the active mode only in response to receiving a user consent.
20. The system of claim 10 , wherein detecting the user activity comprises determining that a user has turned to face the computing device.