The optimal design for VoIP is to minimize network resource competition between VoIP traffic and non-VoIP traffic. Doing this generally includes two approaches, which might be used together or separately.
First, whenever VoIP and non-VoIP traffic compete for bandwidth, VoIP is prioritized over non-VoIP traffic, often such that VoIP traffic has first priority to the bandwidth, although total bandwidth usage might be restricted. The first is done to minimize any queuing latency and/or jitter, the second in case the VoIP traffic somehow begins to run crazy (or its priority is being abused by non-VoIP traffic).
Second, edge VLANs might be dedicated to just VoIP hosts. This to avoid any unnecessary broadcast (or multicast) traffic hitting those hosts which they do not need. Such VLANs also allow security filtering for traffic flowing in/out. This to further assure those voice VLAN only carry traffic supporting VoIP hosts. (BTW, otherwise, there's really nothing technically special about a "voice" VLAN.)
Ideally, VoIP hosts, should have their own dedicated port, but often the edge port is shared with a data host (using the VoIP host as mini VLAN enabled switch). Such sharing on the edge switch is often done by configuring the edge host as a (Cisco) trunk port or as a (Cisco) access port with a "voice" VLAN also assigned to it. (The latter is just a "special" two VLAN trunk.)
One issue with "shared" edge ports, although the port's data and voice VLANs are logically separate, they do share the bandwidth going to the VoIP host. I.e. the VoIP host can still be impacted by data frames consuming bandwidth on the edge port. (For such a shared edge port, like other ports where VoIP and non-VoIP share bandwidth, prioritization for VoIP traffic is recommended.)
I'll mention, brand A VoIP phones, design guides, recommend only half as many VoIP hosts be placed within a VoIP VLAN where ports are shared with data hosts.
