Phoneme Graph Segmentation for Speech Recognition Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current LVCSR and NLU technologies face challenges in achieving accurate and efficient human-machine speech communication, particularly in dynamically changing data source content, leading to high computational complexity in speech recognition.
Innovation Solution
The method involves parsing a data source to obtain partial word lists, compiling them into phoneme graphs, combining these graphs, and using a speech recognizer to improve recognition efficiency by reducing computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If dynamic grammar generation is performed according to data source content changes, then adaptability is improved, but computational complexity increases
Solution Approach 1:
The patent segments the data source into multiple parts and generates grammar and phoneme graphs for each segment independently. This allows the system to handle dynamic content changes by processing only the affected segments rather than regenerating everything, thus reducing computational complexity while maintaining adaptability.
Solution Approach 2:
The patent pre-compiles phoneme graphs for different data source contents before actual speech recognition occurs. By preparing these phoneme graphs in advance, the system avoids complex real-time grammar generation during speech recognition, thereby reducing computational complexity while maintaining the ability to adapt to content changes.
2Measurement precision
If complete grammar generation is performed for the entire data source, then recognition accuracy is improved, but recognition efficiency deteriorates
Solution Approach 1:
The patent divides the data source into segments and generates phoneme graphs for each segment separately. During speech recognition, only the relevant segment phoneme graphs are activated and combined, rather than processing the entire grammar tree. This maintains recognition accuracy for the relevant content while significantly improving recognition efficiency by reducing the overall computational burden.
Solution Approach 2:
The patent generates complete phoneme graphs for all possible data source contents in advance (excessive action), but only activates and combines the necessary portions during actual speech recognition (partial action). This approach ensures recognition accuracy for any possible input while improving efficiency by avoiding unnecessary processing of irrelevant content.
3Adaptability or versatility
If phoneme graphs are combined for all possible data source contents, then adaptability is improved, but computational complexity increases
Solution Approach 1:
The patent pre-compiles phoneme graphs for all possible data source contents in advance, storing them for later use. During actual speech recognition, the system only combines the necessary phoneme graphs corresponding to the current data source content, rather than processing all possible contents. This maintains adaptability to any data source content while reducing computational complexity by avoiding unnecessary combinations.
Solution Approach 2:
The patent generates phoneme graphs for all possible data source contents (excessive action) but only combines and processes the relevant portions during actual speech recognition (partial action). This approach ensures the system can adapt to any content while minimizing computational complexity by avoiding unnecessary processing of irrelevant content.
Data Source
AI summary
The present invention discloses converting a text form into a speech. In the present invention, partial word lists of a data source are obtained by parsing the data source in parallel or in series. The partial word lists are then compiled to obtain phoneme graphs corresponding, respectively, to the partial word lists, and then the obtained phoneme graphs are combined. Speech recognition is then conducted according to the combination results. According to the present invention, computational complexity may be reduced and recognition efficiency may be improved during speech recognition.