The Research And Realization Of Speech Enhancement Algorithm In VoIP System

In recent years, with the popularity of the Internet and VoIP (Voice over IP)technology, internet voice transmission technology has been more popular. In VoIPsystem, bad voice quality problems is ubiquitous, especially in the case of the largenoise around, which makes VoIP systems difficult to be widely used. Therefore, thedesign of a speech enhancement technology for VoIP system is imminent. The mainpurpose of speech enhancement technology is to decrease the noise in noisy speech,enhanced voice and eventually extract voice without noise. This research choose thespeech enhancement algorithm based on short-time spectral estimation, and thenadding the AGC (Automatic Gain Control) module to optimize it, making the inputvoice more clear and understandable in VoIP network.This paper researched the effect of noise reduction by voice codec technology,and using different voice signal within different noise intensity to simulate. Thenchoosing a enhancement algorithms which can fit VoIP system characteristics, andimplementing it in VC project. Afterward, adding the AGC module to optimize.In project implementation, this project designed the algorithm code structuralwhich fit the features of the ARM Cortex-A9processor architecture and optimizingit from the C level and assembly level. The code space, data space and instructionspeed of optimized code reach the embedded operation target. The softwaredevelopment environment is RVDS environment and the GCC environment.In experiment, using the white noise and colored noise to simulate algorithmsand evaluate speech enhancement effect. From the simulation results, conclude thatvoice codec have certain ability to suppress noise, but if the background noise islarge, noise suppression effect can hardly be seen. Using short-time spectrumestimation algorithm, the background noise is significantly reduced, but also hurtthe voice signal. After adding AGC module, the speech energy can be improved.Finally, the noise signal decreased by over80percent based on little damaging ofthe voice signal.