Linear Scoring for Low Power Wake on Voice Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current key phrase detection systems in low resource environments face challenges with robustness due to complexity, leading to issues with noise and false accepts, and do not provide high-quality detection.
Innovation Solution
The implementation of linear scoring procedures for key phrase detection using vectorized operations, which include generating acoustic score vectors and updating state scores to efficiently detect key phrases, optimizing computational efficiency and power usage through SIMD instructions and hardware implementation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Gaussian mixture models are used to model acoustics of variations, then detection accuracy is improved, but computational complexity and power consumption increase
Solution Approach 1:
The patent extracts only the essential acoustic features (MFCCs) from the full audio signal and uses a simplified HMM architecture with limited Gaussian components. This selective extraction maintains detection accuracy while significantly reducing computational complexity for implementation in low-resource environments.
Solution Approach 2:
The patent changes the model parameters by using a reduced number of Gaussian components and simplified HMM structure compared to traditional GMMs. This parameter simplification allows the system to maintain acceptable detection accuracy while operating within low power and computational resource constraints.
2Use of energy by stationary object
If simpler techniques are used in low resource environments, then power consumption is reduced, but robustness to noise and false accepts deteriorates
Solution Approach 1:
The patent incorporates feedback mechanisms through continuous monitoring of acoustic scores and dynamic threshold adjustment. The system uses feedback from the acoustic model outputs to adaptively manage wake states, improving robustness to noise while maintaining low power consumption through efficient state management.
Solution Approach 2:
The patent creates a composite detection system combining multiple components: acoustic feature extraction, HMM modeling, wake word detection, and feedback control. This composite approach integrates the strengths of each component to achieve both low power consumption and high robustness to noise simultaneously.
3Reliability
If complex models are implemented, then detection accuracy is improved, but processing time and memory requirements increase
Solution Approach 1:
The patent segments the detection process into distinct stages: acoustic feature extraction, HMM scoring, wake word matching, and feedback control. This segmentation allows each component to be optimized independently, reducing overall processing time while maintaining detection accuracy through coordinated operation of simplified modules.
Data Source
AI summary
Key phrase detection techniques for applications such as wake on voice are discussed include performing a vectorized operation on a multiple element acoustic score vector for a current time instance including a single state rejection model score and scores for a multiple state key phrase model and a multiple element state score vector for a previous time instance including a previous state score for the single state rejection model and previous state scores for the multiple state key phrase model to generate a multiple element score summation vector and a second vectorized operation on the multiple element score summation vector to determine a multiple element state score vector for the current time instance. The multiple element state score vector for the current time instance may then be evaluated to determine whether received audio input includes a key phrase corresponding to the multiple state key phrase model.