|
Problem
Full duplex hardware enable recording and playing at the same time. This configuration causes echo effect:
- Acoustic echo: Sound that is played by loud speakers is recorded by the microphone.
- Electrical echo: An echo may occur in hybrid connection of full duplex telephone wires.
Problem
Echo suppression:
- The module identifies echo and suppresses it by means of gain control.
- Echo suppression prevents howling effect (positive feedback loop).
- The module allows one party to interrupt the other party's speech.
- The algorithm adapts to background noise level and to echo gain and delay.
- The CPU demands less (uses than 5% of Pentium 100 Mhz CPU).
Echo cancellation:
- The module removes the echo by means of adaptive filtering. It uses a highly effective proprietary algorithm.
- Both parties may talk freely at the same time. The algorithm is stable in double-talk situation (no double-talk detector used).
- Echo cancellation prevents the howling effect (positive feedback loop).
- The module runs on less than 20% of Pentium 100 Mhz CPU.
Combination of echo suppression and cancellation:
- Echo suppression is executed after echo cancellation, and suppresses the residual echo.
The module minimally affects the signal when echo cancellation performs perfectly.
Native implementation:
- Windows 95 DLL runs in ring 3.
- Re-entrant code (written in C).
- The module is activated by a set of APIs.
- The module supports various features for the following modes:
- Hardware half duplex: intelligent switching mechanism.
- Head-set mode: gain control.
- Full duplex echo suppression.
- Full duplex echo cancellation.
- Full duplex combination of echo suppression and cancellation.
- Additional mechanisms:
- Speaker and microphone gain settings.
- Automatic Gain Control.
- Duplicity control for echo suppression.
- Comfort noise for echo suppression.
- Performs well in a real-time environment when activated by a Windows application.
Demo
The following signal demonstrates the combined suppression + cancellation speaker phone that deals with pure echo and double-talk scenarios. The files are in WAV audio format sampled at 8 Khz.
| |
|
