PSTN

The name is historical. The core has been digital since the introduction of time-division multiplexing, and most long-haul transport is now IP, but the abstraction the network presents is unchanged: dial a number in a global numbering plan and reach the endpoint assigned to it, regardless of which of thousands of interconnected carriers serves either end.

The audio constraint is the one that shapes voice AI most. Traditional PSTN voice is sampled at 8 kHz and encoded with G.711 or a lower-bitrate codec, giving a passband of roughly 300 to 3400 Hz. That was chosen for intelligibility of human speech at minimum bandwidth. It removes exactly the high-frequency energy that speech recognisers use to distinguish fricatives, which is why telephony ASR is a distinct problem from ASR generally.

Reaching the PSTN from software means going through a carrier or a CPaaS provider, which brings number provisioning, per-minute charges, interconnection agreements, emergency-calling obligations, and jurisdiction-specific rules on who may be called and when. Those obligations attach to the number and the calling party, not to the software making the call.