Packet Classification for Accurate Voice Quality Estimation

The patent segments packet loss into different categories based on temporal position (e.g., initial packets, middle packets, tail packets) and applies different weighting factors to each segment. This segmentation allows the system to account for the varying impact of packet loss on perceptual quality, thereby improving measurement precision without losing critical information about when packets are lost.

The patent applies local quality by assigning different importance weights to packet losses occurring at different positions in the audio stream. For example, packet losses at the beginning or end of a frame may be weighted differently than losses in the middle, reflecting the local perceptual impact. This approach enhances measurement accuracy by capturing spatial/temporal variations in packet loss impact.

The patent performs preliminary action by pre-defining weighting factors and classification rules for different packet loss scenarios before actual quality measurement. These pre-established parameters (e.g., weighting factors for different packet positions) are stored and applied during measurement, reducing real-time computational complexity while maintaining high measurement precision.

The patent changes parameters by introducing weighting factors that modify the contribution of different packet loss events to the overall quality metric. By adjusting these parameters based on packet position and type, the system achieves higher measurement precision without proportionally increasing system complexity, as the parameter changes are applied through straightforward mathematical operations.

The patent segments the audio stream into distinct temporal regions (e.g., attack region, sustain region, release region) and assigns different weighting factors to each segment. This segmentation enables the system to differentiate packet loss impact by position, improving measurement precision while keeping the classification system relatively simple through clear temporal boundaries.

The patent performs preliminary classification of packets into categories based on their temporal position and importance before quality measurement. By pre-categorizing packets (e.g., marking which packets are critical vs. non-critical), the system reduces the complexity of real-time analysis while maintaining high measurement precision through the use of pre-established classification rules.