We wrote a one way VoIP server and client which performs better under lossy conditions by degrading more gracefully: it fills in dropped time slices with predicted sound
Originally it used a deep model, but that wasn't fast enough for real time; so now it uses some crazy signal processing (this part made by Ishaan)
If time permits, will add simple compression (currently planning to transmit uncompressed audio time slices), erasure codes, and throughput / packet drop forecasting
Log in or sign up for Devpost to join the conversation.