Audio Front End Adaptive Gain for Wakeword Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electronic devices face challenges in accurately detecting voice commands and performing device arbitration due to varying output levels across different devices, leading to inconsistent wakeword energy detection and incorrect selection of the device closest to the user.
Innovation Solution
Implementing an Audio Front End (AFE) component that generates both a fixed gain output for device arbitration and an adaptive gain output for improved wakeword detection, using start and end times of wakeword energy to determine a unique adaptive gain value for each device, ensuring consistent output levels and accurate device selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If output levels are calibrated to desired level for wakeword detection, then wakeword detection accuracy is improved, but device arbitration fails to select the device closest to the user
Solution Approach 1:
The patent segments the audio processing pipeline into two distinct output paths: a first output path calibrated to a desired output level for wakeword detection, and a second output path that preserves varying energy levels for device arbitration. This segmentation allows each path to serve its specific function optimally without interference from the other calibration requirements.
Solution Approach 2:
The patent applies different quality characteristics to different parts of the processing system. The first output is optimized with consistent calibration for wakeword detection sensitivity, while the second output maintains variable energy characteristics that reflect the actual physical distance and acoustic environment, enabling reliable device arbitration.
2Reliability
If output levels are calibrated to have consistent levels for device arbitration, then device selection is improved, but wakeword detection performance degrades
Solution Approach 1:
The patent creates separate processing branches that diverge after the audio front end. One branch applies calibration for device arbitration, while the other branch maintains the uncalibrated or differently calibrated signal for wakeword detection, allowing both functions to operate with their respective optimal settings.
Solution Approach 2:
The patent introduces an intermediary processing stage that generates multiple outputs from a single audio input. This intermediary layer allows the system to provide both the consistently calibrated output needed for device arbitration and the variable-energy output needed for accurate wakeword detection, acting as a mediator between conflicting requirements.
3Reliability
If devices have varying output levels, then device arbitration can select closest device, but wakeword detection becomes inconsistent across devices
Solution Approach 1:
The patent divides the output into two distinct streams: one that maintains varying energy levels to preserve information about device position and acoustic environment for arbitration, and another that applies consistent calibration for stable wakeword detection performance across all devices.
Solution Approach 2:
The patent changes the calibration parameter differently for different output paths. The first output path applies a calibration parameter that ensures consistent output levels for wakeword detection, while the second output path either applies no calibration or a different calibration that preserves the natural variations in energy levels needed for device arbitration.
Data Source
AI summary
A system configured to perform audio processing with adaptive multi-stage output gains. For example, an Audio Front End (AFE) component may generate a first output using a fixed gain value in order to improve device arbitration and a second output using an adaptive gain value in order to improve wakeword detection. A wakeword engine may process the second output to determine that a wakeword is present along with start/end times of the wakeword. The AFE component can use the start/end times to determine an amount of wakeword energy represented in the first output, which is sent to a remote device for device arbitration. The AFE component can also use the start/end times to determine an amount of wakeword energy represented in the second output, which can be used to determine the adaptive gain value that is unique to the device.