Voice Command Model Training via N-gram Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems face challenges in efficiently training models to recognize diverse voice commands, as they require extensive resources and struggle with parsing multiple possible phrases correctly with a limited number of examples.
Innovation Solution
The method involves identifying semantically relevant and irrelevant n-grams in command sentences, removing irrelevant ones from training data, and generating command models that allow for the insertion of irrelevant n-grams at runtime to improve recognition accuracy across actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If extensive training data with multiple phrases is used to improve recognition accuracy, then semantic coverage is improved, but resource intensity and training complexity increase
Solution Approach 1:
The patent segments training data into semantically relevant n-grams and semantically irrelevant n-grams. Command models are trained separately on relevant n-grams for each action, while irrelevant n-grams are handled through insertion operations at runtime. This segmentation allows the system to achieve broad semantic coverage without proportionally increasing training complexity, as the irrelevant portions are managed through a different mechanism (insertion) rather than extensive training examples.
2Adaptability or versatility
If more command examples are provided to cover diverse voice commands, then semantic coverage is improved, but training resource intensity increases
Solution Approach 1:
The patent performs preliminary identification and removal of semantically irrelevant n-grams from training data before model training. By pre-processing the training data to extract only relevant n-grams for each action, the system reduces the volume of training data needed while maintaining the ability to handle diverse commands. The irrelevant n-grams are accounted for through insertion operations during runtime rather than requiring extensive training examples.
3Adaptability or versatility
If traditional speech recognition training is used to recognize diverse commands, then command coverage is improved, but resource intensity increases
Solution Approach 1:
The patent extracts and removes semantically irrelevant n-grams from the training data for each command model. By taking out the irrelevant portions that would otherwise require extensive training to handle correctly, the system reduces training resource intensity. The extracted irrelevant n-grams are then handled through insertion operations at runtime, allowing broad command coverage without proportionally increasing training resources.
Data Source
AI summary
A method includes accessing data specifying a set of actions, each action defining a user device operation and for each action: accessing a corresponding set of command sentences for the action, determining first n-grams in the set of command sentences that are semantically relevant for the action, determining second n-grams in the set of command sentences that are semantically irrelevant for the action, generating a training set of command sentences from the corresponding set of command sentences, the generating the training set of command sentences including removing each second n-gram from each sentence in the corresponding set of command sentences for the action, and generating a command model from the training set of command sentences configured to generate an action score for the action for an input sentence based on: first n-grams for the action, and second n-grams for the action that are also second n-grams for all other actions.