Voice Identification Confidence Threshold Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current voice identification technologies, such as voice assistant software, perform poorly in noisy environments, with noise reduction methods offering limited improvements and sometimes lowering identification rates.
Innovation Solution
A voice identification method and apparatus that dynamically adjust a confidence threshold based on the noise scenario, obtained through noise type and magnitude analysis, using frequency cepstrum coefficients and Gaussian mixture models, to enhance identification accuracy in noisy conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a fixed confidence threshold is used for voice identification, then the identification process is simple and fast, but the identification rate decreases in noisy environments
Solution Approach 1:
The patent applies dynamics by transforming the fixed confidence threshold into a dynamic threshold that adapts to different noise scenarios. The system determines noise scenarios in real-time and adjusts the threshold accordingly, making the identification system flexible and adaptive to changing environmental conditions rather than relying on a static threshold value.
Solution Approach 2:
The patent changes the parameter of the confidence threshold based on the determined noise scenario. By adjusting the threshold parameter dynamically according to noise levels and types, the system optimizes identification accuracy for different acoustic environments without requiring complex reconfiguration of the entire identification pipeline.
2Reliability
If noise reduction processing is applied to improve identification rate, then some noise is reduced, but the improvement effect is not distinct and may lower identification rate
Solution Approach 1:
Instead of applying noise reduction processing to the voice data before identification, the patent inverts the approach by using the noise scenario determination to adjust the confidence threshold. This alternative strategy avoids potential distortion from noise reduction while still improving identification reliability by adapting the decision criterion to match the noise conditions.
3Measurement precision
If a high confidence threshold is used to ensure accuracy, then mis-determination is reduced, but identification rate decreases in noisy environments
Solution Approach 1:
The system dynamically adjusts the confidence threshold based on the determined noise scenario, allowing the threshold to be high when noise is low (ensuring accuracy) and lower when noise is high (maintaining identification rate). This dynamic adaptation resolves the trade-off between accuracy and productivity by making the threshold flexible rather than fixed.
Data Source
AI summary
Embodiments of the present invention provide a voice identification method, which includes: obtaining voice data; obtaining a confidence value according to the voice data; obtaining a noise scenario according to the voice data; obtaining a confidence threshold corresponding to the noise scenario; and if the confidence value is greater than or equal to the confidence threshold, processing the voice data. An apparatus is also provided. The method and apparatus that flexibly adjust the confidence threshold according to the noise scenario greatly improve a voice identification rate under a noise environment.