The design and implementation of acoustic echo cancellation subsystem for multi-party videoconferencing systems

The voice quality of videoconferencing system is seriously affected by acoustic echoes. AEC (acoustic echo cancellation) itself is delicate and complex. It becomes more difficult to cancel acoustic echoes with pure software in an Internet multi-party videoconferencing scenario because of the varying operating system time-delay, complex echo path and network delay jitters. We propose a novel effective and adaptive software AEC subsystem for multi-party videoconferencing systems. By analyzing the characteristics of echo path delay we originally propose a system time-delay estimator and a reference signal delay adjuster to reduce AEC computation amount and to improve its real-time performance. The AEC subsystem is fully implemented, embedded into a new multi-party videoconferencing system (DVTS Plus) and then tested in a real conference scenario. Results show that our proposed subsystem could provide good conference experience with echo free voice in high quality for all users in a multi-party videoconference with small additional amount of computation. The average value of ERLE could reach 50 db during double-talk and it provides a perfect real time performance. Thanks to the computation amount decrease, our proposed AEC subsystem could work perfectly in larger conference room compared to other existing commercial AEC solutions.