You'll need to dual-home them using layer 3 interfaces. You'll configure the two Ethernet interfaces on the VG224 in two different subnets and attach one link to each switch. Then you need to create a Loopback address on the VG224, have it participate in your IGP (EIGRP or OSPF) and announce that address. Optionally you can crank down the timers for your IGP's hello and dead intervals. Then you bind your signaling and control to that Loopback interface. The method for doing that varies depending on whether you're using H.323, MGCP or SCCP and you didn't mention which.
The best way to handle interface config and subnet allocation depend on what sort of VG224 density you have on each switch (are they spread out around a campus or is it a stack of VG224s going into one switch?) and what model switches you have. In general you might use "no-switchport" /30 subnet interfaces on IOS-based switches coupled to one or a few VG224s, or a larger VLAN-based subnet if you have a high density of VG224s connecting to one switch (or if you're using CatOS switches and you don't have a choice).
When the VG224 first started shipping, it supported OSPF but not EIGRP. I ran into that when I tried to deploy them in a redundant fashion like yours, in an environment where EIGRP was already prevalent. I had to work around it with route redistribution. I worked with a friend in TAC to get a bug filed, and EIGRP is now available on the VG224 as of IOS 12.3(13)T and later.
At the customer sites where we've done this (including the IGP timer tuning), we were able to pull out Ethernet links and carry on existing conversations with no more than a second or two interruption.
