Voice Onset Detection for Mixed Reality Speech Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems in mixed reality environments face challenges in minimizing power consumption and improving accuracy, particularly when using wake-up word systems that continuously listen for a trigger phrase, leading to false positives, false negatives, and power inefficiency.
Innovation Solution
A method that involves receiving audio signals via microphones, determining if they contain a voice onset event, and subsequently waking processors to identify a predetermined trigger signal. This approach selectively activates the speech processing engine only when a valid trigger signal is detected, reducing unnecessary processing and power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the wake-up word system continuously listens for a trigger phrase, then the device can respond to voice commands, but power consumption increases and false positives occur
Solution Approach 1:
The wake-up word detection system is divided into multiple processing stages with different processors. The first processor continuously monitors for wake-up words at low power, while the second processor is activated only after voice onset detection to perform comprehensive speech recognition. This segmentation allows the system to maintain reliability for wake-up word detection while significantly reducing overall power consumption by keeping the more power-intensive processing components inactive during normal operation.
Solution Approach 2:
The system performs preliminary voice onset detection using a lightweight first processor before activating the full speech processing engine. By detecting voice onsets in advance and using this information to predict whether full speech processing is needed, the system can avoid unnecessary activation of power-intensive components while maintaining accurate wake-up word response.
2Measurement precision
If the speech processing engine is continuously active, then speech recognition accuracy is maintained, but power consumption increases
Solution Approach 1:
The speech processing engine operates dynamically rather than statically - it transitions between inactive and active states based on real-time voice onset detection. The first processor continuously monitors for voice onsets and only activates the second processor when voice activity is detected, creating a dynamic system that adapts its processing state to current conditions. This dynamic operation maintains speech recognition accuracy when needed while minimizing power consumption during idle periods.
Solution Approach 2:
The first processor acts as an intermediary between the microphone and the second processor. It receives audio signals, detects voice onsets, and uses this information to control activation of the second processor. This intermediary role allows the system to filter out unnecessary processing requests and only activate the full speech processing engine when voice activity is detected, maintaining accuracy while reducing power consumption.
3Ease of operation
If the device uses wake-up word systems, then voice command activation is enabled, but false positives and false negatives occur
Solution Approach 1:
The system incorporates feedback mechanisms where the first processor continuously monitors audio signals for voice onsets and uses this feedback to control activation of the second processor. When voice onset is detected, the system activates comprehensive speech processing to confirm the wake-up word. This feedback loop allows the system to respond reliably to voice commands while using the feedback information to distinguish between actual wake-up words and false triggers, reducing both false positives and false negatives.
Data Source
AI summary
Disclosed herein are systems and methods for processing speech signals in mixed reality applications. A method may include receiving an audio signal; determining, via first processors, whether the audio signal comprises a voice onset event; in accordance with a determination that the audio signal comprises the voice onset event: waking a second one or more processors; determining, via the second processors, that the audio signal comprises a predetermined trigger signal; in accordance with a determination that the audio signal comprises the predetermined trigger signal: waking third processors; performing, via the third processors, automatic speech recognition based on the audio signal; and in accordance with a determination that the audio signal does not comprise the predetermined trigger signal: forgoing waking the third processors; and in accordance with a determination that the audio signal does not comprise the voice onset event: forgoing waking the second processors.