During unfavorable network conditions, a system provides a fallback mode that ensures conversation intelligibility and eliminates dropouts. In the fallback mode, the system uses data retransmission and mitigates retransmission delay by breaking input and output audio into segments for flow control and by involving the users in flow control. On the receive end, when necessary, the system pauses audio output and indicates the pause to the receiving user to inform the receiving user that the pause is a system artifact rather than silence from the transmitting user. On the transmit end, when necessary to avoid accumulation of excessive end-to-end delay, the system suspends audio input and indicates the suspension to the transmitting user to inform the transmitting user that audio input is temporarily being ignored. The transmit end of the system decides when to suspend audio input based on current end-to-end delay, which is continually communicated from the receive end.