Audio Coding of Spectral LSBs Under Bitrate Constraints

The patent applies local quality by differentiating the encoding treatment of spectral values based on their individual characteristics. Specifically, it encodes least significant bits for some spectral values while omitting them for others, creating non-uniform local quality across the spectrum. This allows the system to allocate bitrate resources selectively to spectral values that benefit most from higher precision, thereby improving overall audio quality while controlling total bitrate consumption within the bit budget.

The patent changes the parameter of quantization precision by selectively encoding least significant bits for different spectral values. Instead of applying a uniform global-gain that increases distortion, the system varies the precision parameter locally across different spectral values. This allows maintaining lower overall distortion while still reducing bitrate consumption, as the precision is adjusted per-spectral-value rather than globally, thereby improving audio quality compared to traditional global-gain approaches.

The patent inverts the traditional truncation approach by selectively retaining least significant bits for certain spectral values while omitting them for others, rather than uniformly truncating high-frequency coefficients. This inverted strategy allows the system to maintain better audio quality at high bitrates by preserving important fine-details in selected spectral regions, while still achieving bitrate compliance through selective omission. The approach reverses the conventional wisdom that high-frequency truncation is necessary, instead using selective precision reduction across the entire spectrum.

The patent extracts the least significant bit encoding decision from the global rate-loop optimization process and handles it separately on a per-spectral-value basis. By taking out this specific encoding decision from the complex iterative global optimization, the system avoids the computational burden of multiple rate-loop iterations while still achieving optimal or near-optimal bitrate allocation. The selective LSB encoding can be determined through simpler local criteria, thereby improving audio quality without proportionally increasing computational complexity.