Transform Coder Scale Factor Weighting for Speech Quality

The patent applies local quality by differentiating the treatment of scale factors based on their magnitude. Small-scale factors (corresponding to spectral valleys) are encoded with higher precision using more bits, while large-scale factors (corresponding to spectral peaks) are encoded with fewer bits. This localized differentiation of encoding precision resolves the contradiction by allocating bits according to perceptual importance rather than uniformly, thereby maintaining speech quality in critical regions while reducing overall bit rate.

The patent changes the parameter of bit allocation from a fixed uniform value to a dynamic value that varies with the magnitude of scale factors. By introducing a bit allocation mechanism that adapts to the local spectral characteristics (changing the parameter based on scale factor magnitude), the system achieves both lower overall bit rate and maintained speech quality in perceptually critical regions.

The patent implements local quality by assigning different weights to different Bark scales based on their spectral significance. Instead of a uniform weight function, the system calculates weights that reflect the local importance of each spectral band, allowing accurate representation of significant spectral features while simplifying the encoding of less important regions.

The patent introduces dynamics by making the weight function adaptive rather than static. The weights are calculated based on the actual spectral characteristics of the input signal, allowing the encoding system to dynamically adjust its precision allocation according to the signal's local features, thereby achieving both accuracy and efficiency.