Local Speech Classifier Using Sensor-Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computing devices face challenges in providing accurate speech recognition without relying on cloud-based servers, especially when cellular wireless coverage is limited or unavailable, leading to delays and increased computational complexity.
Innovation Solution
The implementation of classifiers that map text strings to actions based on string-features and sensor-features, reducing the need for complex semantic analysis and allowing the device to autonomously perform speech recognition with reduced computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If cloud-based servers are used for speech recognition, then recognition accuracy is improved, but device reliability deteriorates when cellular wireless coverage is limited or unavailable
Solution Approach 1:
The patent introduces classifiers as intermediary components that bridge the gap between simple local processing and complex cloud-based recognition. These classifiers are trained offline using labeled data and serve as local decision-making agents that can accurately recognize speech patterns even without cloud connectivity, thus maintaining reliability while preserving reasonable recognition accuracy
Solution Approach 2:
The system performs preliminary training of classifiers offline using labeled speech data before actual use. This preliminary action creates pre-configured recognition models that can be deployed on devices without requiring cloud connectivity during operation, enabling the device to reliably recognize speech independently
2Measurement precision
If cloud-based servers are used for speech recognition, then recognition accuracy is improved, but response time deteriorates due to delays
Solution Approach 1:
Classifiers act as local intermediaries that eliminate the need for cloud communication during speech recognition operations. By processing speech locally through pre-trained classifiers, the system removes network transmission delays while maintaining accurate recognition, thus resolving the time-accuracy tradeoff
3Measurement precision
If complex semantic analysis is used for speech recognition, then recognition accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the speech recognition task into two distinct phases: offline training phase where complex learning occurs to create classifiers, and online recognition phase where simple classification operations are performed. This segmentation moves computational complexity to the offline phase, enabling accurate recognition with minimal runtime computational requirements
Solution Approach 2:
The system creates simplified copies of speech recognition capabilities in the form of trained classifiers that can be deployed on resource-constrained devices. These classifier copies capture the essential recognition patterns without requiring the full computational infrastructure needed for complex semantic analysis
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed embodiments relate to mapping an utterance to an action using a classifier. One illustrative method involves receiving, by a computing device, an audio utterance; determining a text siring based on the utterance; determining a string-feature vector based on the text siring; receiving sensor data; determining a sensor- feature vector based on the sensor data: selecting a target classifier from a set of classifiers, wherein the target classifier is selected based on a determination that a string-feature criteria of the target classifier corresponds to at least one string-feature of the string-feature vector and a sensor-feature criteria of the target classifier corresponds to at least one sensor-feature in the sensor-feature vector; and initiating, by the computing device, a target action that corresponds to the target classifier.