Distributed Keyword Detection via Local and Remote NLU Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice-assisted media playback systems face challenges in accurately detecting voice commands without transmitting audio recordings to the cloud, leading to privacy concerns and inefficiencies in processing voice inputs.
Innovation Solution
Implementing a network microphone device with a command-keyword engine that detects and processes voice inputs locally, using a command-keyword engine in parallel with a wake-word engine, which allows for quicker responses and increased user privacy by performing actions directly without cloud processing, and utilizing multiple network microphone devices for cross-checking and expanding keyword libraries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If voice inputs are transmitted to the cloud for processing, then keyword detection accuracy can be improved through comprehensive processing, but user privacy is compromised and processing time increases
Solution Approach 1:
The system segments keyword detection into two parts: common keywords are detected locally on the playback device for immediate response, while less common keywords are detected by the voice assistant service on remote devices. This segmentation allows privacy-sensitive operations to remain local while still providing comprehensive keyword coverage through the distributed architecture.
Solution Approach 2:
The system introduces an intermediary architecture where the voice assistant service acts as a mediator for keywords not detected locally. The playback device first attempts local detection, and only for undetected keywords does the system involve remote processing through the voice assistant service, thus minimizing cloud transmission while maintaining comprehensive detection capability.
2Adaptability or versatility
If cloud processing is used for keyword detection, then comprehensive keyword coverage is achieved, but processing speed decreases due to network transmission
Solution Approach 1:
The keyword detection function is segmented between local and remote devices. Common keywords with higher detection probability are processed locally on the playback device for immediate response, while less common keywords are handled by the voice assistant service on remote devices. This segmentation optimizes both response speed for frequent keywords and comprehensive coverage for all keywords.
Solution Approach 2:
The system performs partial keyword detection locally without waiting for complete cloud processing. By detecting common keywords locally and only involving remote processing when needed, the system achieves sufficient keyword coverage without the full delay of cloud-based processing for every input.
3Reliability
If multiple network microphone devices are deployed, then keyword detection reliability is improved through cross-checking, but system complexity increases
Solution Approach 1:
The system segments the keyword detection workload across multiple playback devices and the voice assistant service. Each playback device independently detects keywords locally, and the voice assistant service provides additional detection capability for less common keywords. This segmentation improves reliability through distributed detection while keeping individual device complexity low.
Solution Approach 2:
The system merges the keyword detection capabilities of multiple playback devices with the voice assistant service into a unified distributed detection system. This combination allows cross-checking and verification across devices, improving overall detection reliability while sharing the processing burden to avoid excessive complexity in any single component.
Data Source
Figure 1A
Figure 1B
Figure 2A~2B
AI summary
In one aspect, a playback device includes at least one microphone configured to detect a voice input and generate sound input data. The playback device detects a first command keyword in the detected sound and, in response, makes a first determination, via a first local natural language unit (NLU), whether the input sound data includes at least one keyword within a first predetermined library of keywords. The playback device receives an indication of a second determination made by a second NLU that the input sound data includes at least one keyword from a second predetermined library of keywords. The playback device compares the results of the first determination and the second determination and, based on the comparison, forgoes further processing of the input sound data.