Abstract: An apparatus (30) and method minimize voice delay by combining comfort noise with subframe insertions to fill gaps in speech and maintain a minimum delay for packets arriving late. When a packet is late, short durations of sound are repetitively inserted until the late packet arrives. When the late packet finally arrives, the durations of future speech pauses are modified so there is no voice degradation.

Abstract: A digital telecommunication network (10) provides an audio processor (56) for each call in each external node (14). The audio processor (56) includes a vocoder and additional functions. At a source external node (14), the audio processor (56) detects DTMF events and encodes the events in a call-related signalling packet (26) which conveys a history of all DTMF activity at the source node over the past several seconds. In addition, the source node detects silence and configures a call-related signalling packet (26) to instruct a destination node (14) how to synthesize comfort noise. Speech is encoded as vocoded voice packets rather than call-related signalling packets. At a destination external node, the node's audio processor (56) distinguishes call-related signalling from vocoded voice. The audio processor (56) synthesizes voice, comfort noise, DTMF signals, and other signals in response to the call-related signalling packets and vocoded voice packets.