Speech Error Concealment Using Local Quality Adaptation

The patent applies local quality by differentiating the error concealment approach based on the local characteristics of the lost frame. The system determines whether the lost frame contained voiced or unvoiced speech and applies different synthesis methods accordingly - using fundamental frequency signals for voiced segments and noise signals for unvoiced segments. This localized adaptation maintains acoustic quality by matching the statistical properties of the original speech in each specific region.

The patent employs parameter changes by dynamically adjusting the spectral and temporal parameters of the substitute speech signal based on the characteristics of surrounding frames. The system modifies parameters such as fundamental frequency, spectral envelope, and noise variance to match the local speech characteristics. This allows the substitute frames to blend seamlessly with the original speech, maintaining perceptual quality while ensuring continuous output.

The patent uses parameter changes to adapt the substitute speech signal characteristics dynamically. By analyzing parameters such as autocorrelation values, energy levels, and spectral properties of surrounding frames, the system adjusts the synthesis parameters (fundamental frequency, spectral envelope, noise variance) to match the local speech characteristics. This approach maintains high perceptive quality without requiring complex neural networks or iterative optimization algorithms.

The system applies local quality analysis by examining the specific characteristics of the frame preceding the lost frame and the frame following it. Based on this local analysis, the system determines whether to use voiced or unvoiced synthesis methods and adjusts the parameters accordingly. This localized approach provides high quality error concealment with relatively simple processing compared to global optimization methods.