Realistically, it depends on the size of the current spanning tree and how it is disrupted.
In an RSTP or MSTP environment, if an interface goes hard down (eg: it's disconnected or otherwise shut down), then the topology change will trigger immediately - between only two switches, a new tree should establish in less than a second and forwarding will recommence.
If there is break not caused by link down (eg: configuration change, intermediate device failure etc), then RSTP and MSTP will wait for 3x Hello Interval (3x2 (default) = 6 seconds by default) before re-converging.
STP on the other hand is a lot slower because it has a slightly different state machine - it will wait 10x Hello Interval (20 seconds) for a BPDU timeout, and then sits in listening State for another 15 seconds, followed by the learning state for another 15 seconds, giving you somewhere around 50 seconds to converge.
Bear in mind though that these figures are for a single switch - if you have a large network diameter, downstream switches may start detecting failure at slightly later times, meaning they will start this process later and ultimately add more time to a complete topology re-convergence.
Added to this is when you're using protocols like VSTP/PVST/PVST+ the switch needs to do this re-convergence for every VLAN, which if there are a lot can be quite taxing on the CPU thus slowing down the r-econvergence further.