Multi-Channel Comfort Noise Injection for Low-Bitrate Stereo Decoding

The system performs preliminary action by estimating the background noise spectrum during active speech segments and storing this information for use during inactive segments. This advance preparation ensures that when speech is absent, the pre-estimated noise spectrum can be immediately applied to generate continuous comfort noise, preventing the intermittent noise perception that occurs with traditional methods.

The patent introduces an intermediary mechanism - a background noise estimator that processes the decoded mono down-mixed signal to extract noise spectrum information. This estimator acts as a mediator between the compressed audio signal and the comfort noise generation process, enabling the system to derive accurate noise characteristics from the limited bitrate signal and apply them appropriately during inactive segments.

The system applies segmentation by separating the comfort noise injection process into distinct left and right channel processing paths. Instead of treating the audio signal as a single mono stream, the patent processes each stereo channel independently, applying the estimated noise spectrum to generate separate comfort noise signals for each channel. This segmentation restores spatial immersion while maintaining manageable device complexity through efficient processing architecture.

The patent implements local quality by applying different noise injection characteristics to different spatial locations (left and right channels). The comfort noise is generated and injected separately for each channel based on the estimated background noise spectrum, creating a spatially distributed noise field that enhances immersion. This approach ensures that each channel contributes appropriately to the overall spatial audio experience.