Adaptive noise modeling speech recognition system
View Patent ↗An adaptive noise modeling speech recognition system improves speech recognition by modifying an activation of the system's grammar rules or models based on detected noise characteristics. An adaptive noise modeling speech recognition system includes a sensor that receives acoustic data having a speech component and a noise component. A processor analyzes the acoustic data and generates a noise indicator that identifies a characteristic of the noise component. An integrating decision logic processes the noise indicator and generates a noise model activation data structure that includes data that may be used by a speech recognition engine to adjust the activation of associated grammar rules or models.
1. A method of adjusting an aggressiveness of a speech recognition engine grammar rule, comprising:
receiving an acoustic signal;
evaluating the acoustic signal with a processor to identify a source of noise in the acoustic signal;
selecting a first noise model to process the acoustic signal, the first noise model selected based on the source of the noise;
generating a noise indicator indicative of a number of prior occurrences of the identified source of noise in the acoustic signal;
processing the noise indicator with a decision logic processor;
generating a noise model activation data structure, at the decision logic processor, based on the noise indicator, the noise model activation data structure comprising data to adjust a setting of a speech recognition rule applied to the first noise model to recognize an utterance in the acoustic signal.
2. The method of claim 1 , where the noise model activation data structure further comprises data characterizing a statistical analysis of a plurality of prior noise indicators.
3. The method of claim 1 , further comprising using the noise model activation data structure to adjust a speech recognition transition probability setting.
4. The method of claim 3 , where the adjustment to the setting of a speech recognition rule applied to the first noise model to recognize an utterance in the acoustic signal and the adjustment to the speech recognition transition probability setting are different.
5. An integrating decision logic that adjusts an activation of a speech recognition grammar rule, comprising:
a processor executing decision logic configured to perform the operations of:
receiving a noise indicator signal associated with a noise component of an input acoustic signal;
extracting characteristics of the noise component from the noise indicator signal; and
generating a noise model activation data structure based on the characteristics which were extracted, the noise model activation data structure containing a grammar rule adjustment indicator that adjusts how a previously configured rule grammar database of a speech recognition engine recognizes utterance subcomponents of the acoustic signal.
6. The integrating decision logic of claim 5 , where the grammar rule adjustment indicator comprises an adjustment to a transition probability rule.
7. The integrating decision logic of claim 5 , where the grammar rule adjustment indicator comprises an adjustment to a duration probability rule.
8. The integrating decision logic of claim 5 , where the grammar rule adjustment indicator comprises an adjustment to an acceptance threshold rule.
9. The integrating decision logic of claim 5 , where the grammar rule adjustment indicator is formatted to adjust less than an entirety of grammar rules applied by the speech recognition engine.
10. An adaptive noise modeling speech recognizer, comprising:
a processor executing a noise modeler configured to perform the operations of:
receiving an acoustic signal containing a speech component and a noise component;
identifying a characteristic of the noise component;
generating a noise model activation data structure based on the characteristic of the noise component that was identified, the noise model activation data structure containing a grammar rule adjustment indicator that adjusts a rule of a grammar database included in a speech recognition engine used to recognize an utterance subcomponent of the acoustic signal; and
transmitting the received acoustic signal and the noise model activation data structure to the speech recognition engine.
11. The adaptive noise modeling speech recognizer of claim 10 , where the characteristic comprises a noise level of the acoustic signal.
12. The adaptive noise modeling speech recognizer of claim 10 , where the grammar rule adjustment indicator comprises a magnitude of the noise component.
13. The adaptive noise modeling speech recognizer of claim 10 , where the processor is further configured to dynamically update the grammar rule adjustment indicator in real-time.
14. The adaptive noise modeling speech recognizer of claim 10 , where characteristic comprises a type of noise.
15. The adaptive noise modeling speech recognizer of claim 14 , where the grammar adjustment indicator comprises frequency data representing how often the type of noise is actually identified in the acoustic signal.
16. The adaptive noise modeling speech recognizer of claim 10 , where the grammar adjustment indicator comprises a confidence score representing a likelihood of a type of noise.