Audio Quality Assessment via Noise Contrast Parameter
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the quality of audio signals, such as PESQ, fail to accurately predict speech quality in telecommunications systems like Voice-Over-IP due to simplistic noise interpretation, which does not account for noise level variations and their impact on perceived quality.
Innovation Solution
A method that processes audio signals by scaling reference and output signals to a fixed intensity level, calculating a noise contrast parameter based on noise level variations, applying noise suppression in the perceptual loudness domain, and deriving a quality indicator from the difference signal to improve correlation with subjective testing results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current objective measurement methods like PESQ are used to determine speech quality, then the measurement process is efficient and objective, but the accuracy of quality prediction deteriorates due to simplistic noise interpretation that does not account for noise level variations
Solution Approach 1:
The patent changes the parameters used in quality assessment by introducing a noise contrast parameter that considers both noise level and its variations over time. Instead of using fixed threshold values, the method dynamically adjusts parameters based on local signal levels and noise characteristics, thereby improving prediction accuracy while maintaining computational efficiency
Solution Approach 2:
The measurement method transitions from static noise interpretation to dynamic noise contrast analysis. The system continuously adapts to varying noise levels and their temporal variations, making the quality assessment responsive to changing acoustic conditions rather than relying on fixed thresholds
2Measurement precision
If noise suppression is applied in the perceptual loudness domain based on noise contrast parameter, then the correlation with subjective testing improves, but the processing complexity increases
Solution Approach 1:
The patent introduces the noise contrast parameter as an intermediary that bridges objective measurements and subjective perception. This parameter serves as a mediator that captures the perceptual impact of noise variations, allowing the system to achieve better correlation with subjective testing without requiring full-blown subjective evaluation procedures
Solution Approach 2:
The method performs preliminary noise contrast analysis and suppression in the perceptual loudness domain before final quality assessment. By pre-processing the signals to account for noise characteristics, the system simplifies subsequent processing steps while maintaining high correlation with subjective results
Data Source
AI summary
The invention relates to a method for determining a quality indicator representing a perceived quality of an output signal of an audio device with respect to a reference signal. Such audio device may for example be a speech processing system. In the method the reference signal and the output signal are processed and compared. The processing includes dividing the reference signal and the output signal into mutually corresponding time frames. The processing further includes scaling the reference signal towards a fixed intensity level. Time frames of the output signal are selected based on measurements performed on the scaled reference signal. Then, a noise contrast parameter is calculated based on the selected time frames of the output signal. A noise suppression is applied on at least one of the reference signal and the output signal based on the noise contrast parameter. Finally, the reference signal and the output signal are perceptually subtracted to form a difference signal, and the quality indicator is derived from the difference signal.