Speech Recognition End Point Detection for False Activation Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition technologies in electronic devices often suffer from inaccuracies due to passive errors, such as meaningless searches or unintended voice inputs, which degrade user experience and reduce the reliability of voice command systems.
Innovation Solution
Implementing a speech signal-based end point detector and contextual end point detector to identify the end of speech, along with a time limit for voice input, to enhance recognition accuracy by performing additional analysis based on input time and text length, and providing a recognition result only after confirming the intent through secondary analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If speech recognition is activated without time limits or end point detection, then responsiveness to user voice commands is improved, but false activations and unintended operations increase
Solution Approach 1:
The system performs preliminary actions by detecting end points of speech signals and establishing time limits before finalizing recognition results. This preliminary detection of speech boundaries and temporal constraints prevents false activations while maintaining responsiveness to legitimate user commands.
Solution Approach 2:
The system implements feedback mechanisms where recognition results are validated against the detected speech end points and time limits. This feedback loop allows the system to correct or reject recognition results that would lead to false activations, thereby improving reliability without sacrificing responsiveness.
2Measurement precision
If additional analysis operations are performed to confirm recognition errors, then recognition accuracy is improved, but processing time increases
Solution Approach 1:
The system performs preliminary analysis operations to detect speech end points and establish time limits before completing the full recognition process. This preliminary action filters out clearly erroneous recognitions early, reducing the need for extensive additional analysis and thereby minimizing processing time while maintaining accuracy.
Solution Approach 2:
The system applies partial analysis operations selectively - performing full analysis only when necessary based on the detected speech characteristics and time limit violations. This partial action approach avoids excessive processing time for clearly valid or clearly invalid recognitions while maintaining high accuracy for ambiguous cases.
3Reliability
If end point detection and time limits are implemented, then false activations are reduced, but device complexity increases
Solution Approach 1:
The end point detection module serves multiple functions: it detects speech boundaries, determines speech duration, validates recognition results, and enforces time limits. This multi-functionality reduces the need for separate dedicated components, thereby limiting the increase in device complexity while achieving reliable false activation reduction.
Data Source
AI summary
An electronic device includes: a memory storing instructions; and a processor operatively connected to the memory and configured to execute the instructions to: perform a first analysis operation of confirming a recognition error for a voice based on an input time of the voice; determine whether to perform a second analysis operation of confirming the recognition error in response to a result of the first analysis operation; based on a determination to perform the second analysis operation, perform the second analysis operation based on a text corresponding to the voice; and provide a recognition result of the voice based on a result of the second analysis operation.