Real-time high accurate cell loss recovery technique for speech over ATM networks

In speech transmission over asynchronous transfer mode (ATM) networks, ATM cell loss cannot be avoided. Over 0.1 percent of cell loss is detected by the human ear, and over 1 percent of cell loss degrades speech articulation. Therefore, the recovery of cell loss is necessary for the improvement of speech quality. For this improvement, several methods have been studied. However, they have problems of speech quality, the delay acceptable for ATM, or the computational and implementational complexity. Moreover, most of them were not evaluated in a realistic telecommunication environment. For speech quality improvement over ATM networks, a real-time high accurate cell loss recovery method is proposed, where a great simplification and large data reduction are achieved by introducing trinary waveforms and encoding. Evaluations in a realistic telecommunication environment show that the method sufficiently improves the speech quality (90-100 percent correct recovery of ATM cell loss, and 20-40 dB improvement of signal-to-noise ratio)