Wakewordless Voice Quickstarts With Speaker-Verified Keyword Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice-assisted media playback systems face challenges with false positives due to traditional wake-word engines, which are prone to triggering on false wake words and consuming additional resources, and existing local voice input processing methods are inefficient and prone to errors.
Innovation Solution
Implementing a mini-SLU pipeline with a neural architecture that detects specific keywords without full ASR, combined with a wakewordless paradigm, and using speaker profiles to enhance confidence in keyword detection, thereby reducing false positives and improving processing speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If traditional wake-word engines are used, then voice-activated control can be implemented, but false positives occur and additional resources are consumed
Solution Approach 1:
The patent extracts and removes the wake-word triggering mechanism from the system, implementing a wake-wordless paradigm where voice commands are processed directly without requiring a wake word prefix, thereby eliminating false positives associated with wake-word detection
Solution Approach 2:
The patent segments the voice processing pipeline into distinct components: a mini-SLU pipeline for quick keyword detection and a full ASR pipeline for comprehensive speech recognition, allowing efficient handling of voice commands without wake words
2Measurement precision
If full ASR (Automatic Speech Recognition) is used for local voice input processing, then accurate speech recognition can be achieved, but processing efficiency decreases and errors increase
Solution Approach 1:
The patent divides the speech recognition system into two segments: a mini-SLU pipeline that quickly processes common keywords and commands, and a full ASR pipeline that handles more complex speech recognition tasks, optimizing both speed and accuracy
Solution Approach 2:
The patent applies partial ASR action by using the mini-SLU pipeline for rapid keyword spotting and only invoking full ASR when necessary, reducing overall processing time and errors while maintaining accuracy for critical commands
3Ease of operation
If wake-word engines are used, then voice commands can be triggered, but processing speed decreases due to additional wake word detection step
Solution Approach 1:
The patent removes the wake-word detection step from the voice processing pipeline, allowing voice commands to be processed directly without the additional time required for wake word recognition, thereby increasing processing speed
Data Source
AI summary
As noted above, example techniques relate to local voice control. A device may monitoring an input sound-data stream representing sound detected by the one or more microphones for keywords and generate a first keyword detection event corresponding to a voice input when one or more keyword engines detect sound data matching at least one first keyword of the one or more keywords. The device determines whether the second voice input matches a particular predetermined speaker profile of one or more predetermined speaker profiles. Based on (i) generating the first keyword detection event and (ii) determining that the second voice input includes sound data matching the particular predetermined speaker profile, the device performs a particular playback command associated with the at least one first keyword.