Weighted Grammar Segmentation for Speech Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems are unreliable and slow, often frustrating users and being costly to maintain, as they struggle to accurately recognize speech responses beyond the most frequently selected options due to incompatible grammars.
Innovation Solution
The system divides dialog turns into segments based on probable user responses, generates weighted grammars for each segment, and exclusively activates them during the corresponding segment, using historical data to assign probabilities and blend grammars for smooth transitions, allowing for dynamic menu generation and improved recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single weighted grammar is created based on overall selection rates, then recognition accuracy for frequently selected options is improved, but recognition accuracy for less frequently selected options deteriorates due to grammar conflicts
Solution Approach 1:
The patent divides the dialog turn into multiple segments based on timing of probable user responses. Each segment has its own weighted grammar tailored to the specific options likely to be selected in that time window. This segmentation resolves the contradiction by preventing grammar conflicts while maintaining high recognition accuracy for frequently selected options in each segment.
Solution Approach 2:
The patent applies local quality by creating grammars with different weight distributions for different segments. Each segment's grammar is optimized locally for the specific options probable in that time window, rather than applying a uniform grammar across all options. This allows high accuracy for frequent options in each segment without compromising recognition of less frequent options in other segments.
2Measurement precision
If historical data is used to create weighted grammars, then recognition accuracy for common patterns is improved, but adaptability to new or varied user responses deteriorates
Solution Approach 1:
The patent makes the grammar dynamic by exclusively activating different weighted grammars for different segments of the dialog turn. The system adapts to varied user responses by selecting and activating the appropriate grammar for each segment based on timing and probable responses, rather than using a single static grammar. This maintains accuracy for common patterns while adapting to new or varied responses.
Solution Approach 2:
The patent uses historical data to preliminarily determine the most probable user responses for each segment and pre-weights the grammars accordingly. This preliminary action based on historical patterns enables high recognition accuracy for common responses while the segment-based structure allows adaptation to varied responses by activating appropriate pre-prepared grammars for each segment.
Data Source
AI summary
Disclosed herein are systems, computer-implemented methods, and computer-readable media for enhancing speech recognition accuracy. The method includes dividing a system dialog turn into segments based on timing of probable user responses, generating a weighted grammar for each segment, exclusively activating the weighted grammar generated for a current segment of the dialog turn during the current segment of the dialog turn, and recognizing user speech received during the current segment using the activated weighted grammar generated for the current segment. The method can further include assigning probability to the weighted grammar based on historical user responses and activating each weighted grammar is based on the assigned probability. Weighted grammars can be generated based on a user profile. A weighted grammar can be generated for two or more segments. The weighted grammar is weighted based on a user profile which includes of information about a number called from, account information, a time of day, and a date. Exclusively activating each weighted grammar can include a transition period blending the previously activated grammar and the grammar to be activated.