Bandwidth Voice-Switching for Echo Reduction in Videoconferencing

The patent segments the audio frequency spectrum into multiple frequency bins and processes each bin independently. By analyzing the spectral content in different frequency ranges separately, the system can identify and suppress far-end sound in specific frequency regions without affecting other frequencies, thereby maintaining sound source localization accuracy while reducing audible echo.

The patent applies different processing strategies to different frequency regions. Instead of uniform echo cancellation across all frequencies, the system identifies which frequency bins contain far-end sound and applies suppression only to those specific bins. This localized approach preserves the quality of near-end speech in unaffected frequency regions while eliminating echo in problematic regions.

The patent dynamically adjusts the amount of far-end sound suppression applied to each frequency bin based on real-time analysis. The system continuously monitors the spectral content and adapts the suppression level for each frequency region, increasing suppression where far-end sound is detected and reducing or eliminating suppression where near-end speech dominates. This dynamic adaptation resolves the contradiction by optimizing both localization accuracy and echo reduction in different frequency regions simultaneously.

The patent changes the spectral parameters of the audio signal by analyzing and modifying specific frequency bins. By identifying frequency regions where far-end sound dominates and applying targeted suppression to those bins while preserving other frequencies, the system achieves both improved sound source localization and reduced audible echo through parameter-based selective filtering.