Hierarchical Speech Recognition Resolution for Latency and Privacy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cloud-based speech recognition systems face issues with privacy concerns, susceptibility to monitoring, high latency, and dependency on connectivity, as they require speech data to be sent to the cloud for processing.
Innovation Solution
A hierarchical speech recognition system with multiple levels of automatic speech recognition (ASR) engines, where the policy manager determines the appropriate level for processing based on factors like confidence rating, privacy settings, and system constraints, allowing for local processing when possible to maintain privacy and reduce latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If cloud-based speech recognition is used, then speech recognition accuracy is improved, but privacy is compromised and latency increases
Solution Approach 1:
The speech recognition system is segmented into multiple hierarchical levels: on-device ASR engines (mobile ASR, on-device ASR) and cloud-based ASR engines. Each level handles different types of recognition tasks, with simpler models processing straightforward commands locally and more complex models handling difficult speech in the cloud, thereby reducing latency for common tasks while maintaining high accuracy for complex recognition.
Solution Approach 2:
Different ASR engine levels are deployed based on local capabilities and requirements. The system configures which ASR engines run locally on the device versus in the cloud, allowing privacy-sensitive applications to use local processing while maintaining the option to escalate to cloud-based high-accuracy engines when needed.
2Measurement precision
If cloud-based speech recognition is used, then speech recognition accuracy is improved, but privacy is compromised
Solution Approach 1:
The system segments speech processing across multiple security boundaries: local ASR engines process speech entirely on the user's device without external transmission, while cloud-based engines handle only necessary cases with appropriate security measures. This segmentation allows users to maintain privacy for routine commands while leveraging cloud accuracy when needed.
Solution Approach 2:
The system enables local processing of speech data through on-device ASR engines, keeping sensitive speech data local to the user's device. This local capability reduces privacy risks by eliminating the need to transmit speech to the cloud for many common recognition tasks.
3Measurement precision
If cloud-based speech recognition is used, then speech recognition accuracy is improved, but dependency on connectivity increases
Solution Approach 1:
The system segments recognition capabilities across multiple levels of connectivity dependency. Local ASR engines operate independently of network connectivity, providing reliable speech recognition for basic commands offline. Cloud-based ASR engines provide enhanced accuracy when connectivity is available, creating a hierarchical system that maintains reliability across varying network conditions.
4Measurement precision
If higher-level ASR engines are used, then word recognition capability is improved, but power consumption increases
Solution Approach 1:
The system segments speech recognition tasks across multiple ASR engine levels with different power requirements. Simpler on-device ASR engines handle routine recognition tasks with low power consumption, while more powerful cloud-based ASR engines handle complex recognition scenarios when additional computational resources are needed and power is available.
Solution Approach 2:
The system dynamically selects which ASR engine level to use based on real-time conditions including power availability, speech complexity, and accuracy requirements. This dynamic adaptation allows the system to optimize the balance between word recognition capability and power consumption rather than using a fixed high-power engine for all tasks.
Data Source
AI summary
A system, method, apparatus and computer readable medium for hierarchical speech recognition resolution. The method of hierarchical speech recognition resolution on a platform includes receiving a speech stream from a microphone. The speech stream is resolved using a lowest possible level automatic speech recognition (ASR) engine of multi-level ASR engines. The selection of the lowest possible level ASR engine is based on policies defined for the platform. If resolution of the speech stream is rated less than a predetermined confidence level, the resolution of the speech stream is pushed to a next higher-level ASR engine of the multi-level ASR engines until the resolution of the speech stream meets the predetermined confidence level without violating one or more policies.