RFC 3261, the standard that defines SIP, calls it "an application-layer control protocol that can establish, modify, and terminate multimedia sessions," and gives internet telephone calls as an example. On a phone call, SIP messages request the call, report that the far end is ringing, confirm the answer and end the call when someone hangs up. The same document marks a limit: "SIP does not provide services." It supplies building blocks, and a carrier builds a service from them. So SIP is a protocol, and SIP trunking is a service that uses it.
The voices do not travel inside SIP messages. RFC 3261 names the Real-time Transport Protocol, or RTP, among the protocols typically used alongside SIP to carry real-time data. Each call on a trunk therefore has two parts, the signaling that manages it and the audio stream itself. This matters when something goes wrong, because a call can connect correctly and still have no sound in one direction. We follow one call through both parts in how SIP trunking works.
The word trunk is older than SIP. Atlantech's post describes a trunk as a link that carries signals between the nodes of a communications system. For a business, that has meant the link between the phone system and the carrier, as distinct from the line to one person's phone. On a SIP trunk the link is not a dedicated cable. It is an arrangement between two systems across an IP network, in which each side is configured to reach the other and to know which traffic to trust.