Classifying VoIP μ-law Packets in Real-Time

Inhalt:
We present an algorithm, which classifies μ-law coded VoIP packets in real time. It is based on our algorithm calculating frame importance off-line, but uses shorter speech segments, an analysis-by-synthesis approach, and a newly developed perceptual evaluation algorithm called PESQlight. Altogether, we are able to reduce the complexity and algorithm delay significantly. The real-time classification has a correlation of up to R=0.63 compared to the reference, off-line algorithm.