To a communications engineer, voice defines the set of characteristics that is necessary to carry through a network in order to deliver meaningful human oral communications. Of these characteristics, the most important is the frequency range, commonly described as zero to 4000 Hz, but, in practice, filtered so that the frequencies below 300 Hz, and often above 2800 Hz, are not in the transmitted signal.
Several types of delay among the sound elements need to be preserved, or human communications may be impaired. The constant, absolute delay from speaking to hearing should be no greater than 160-200 milliseconds; shorter delays are not perceptible. Longer delays will interfere with the spontaneity of the communication, as the listener waits for the response. Unless the speaker and listener are psychologically prepared to expect longer delays, two-way conversation begins to fail to deliver nuances of meaning. A 600 millisecond one-way delay is the rough outside limit for intelligible two-way communication, and such communication will be of the quality of orders sent, perhaps, from an order-taker to a supply clerk or a maker of fast food hamburgers.
People can tolerate longer delays if they know they are making a transcontinental telephone call. When speaking with astronauts on the Moon, a 3.5 second delay was tolerable given the awareness, by all participants, that the communications were literally taking place outside experience with the world, and that a new set of assumptions had to apply. Nevertheless, no one would expect interactive communications with the 20 minute light-speed delay to Mars.
Variability of delay, within a conversation, is quite possibly more important to minimize than absolute delay. Pauses in speech convey meaning. Contrast the difference among "John is a very nice man" and "John is a very [noticeable delay] nice man", and "John is a very nice [noticeable delay] man."
The characteristics that must be preserved present engineering constraints to such technologies as voice over Internet Protocol (VoIP).