WR-RAC: Waveform Regenerator in Real time for Audio Call

In Real time communication audio data packets transmitted via UDP are prone to loss in the Network. These losses results in audio quality degradation in form of glitches, robotic sound or breaks in audio signal on the receiver side. This raises the need for Packet Loss Concealment (PLC) Algorithms for speech waveform recovery in real time, which are efficient in terms of audio quality, power consumption and the stricter latency requirement. We have proposed Waveform Regenerator in Real time for Audio Call (WR-RAC) system for the regeneration of missing speech waveform. The proposed approach uses the Cascaded Network for waveform prediction and has less memory footprint and low inference latency. It also consumes minimal power, which is well suited for real time applications on embedded devices. Waveform Edge Smoothening (WES) post processing module is introduced to limit the effect of discontinuity at the edges between the actual received frame and the next concealed frame.