The version and other parameters that could be used for signaling are generally set by the work producer, i.e. your own node or the pool you are mining with. That will issue the work for your mining hardware and include what block version to produce.
For your own pool or node, you would have to configure your pool or node software to produce work with your desired block version. It depends on the software – it may have some setting that you can set, or you may need to use a different version of the software itself.
If you mine with a third party pool, you’d generally have to switch to a different pool if you wanted to signal.
Note that it is not voting, but rather it is signaling readiness. The purpose is for miners to signal to every other node on the network that they are ready to enforce the new soft fork rules. Since the hashrate hardware is not receiving and validating new blocks and transactions, having it set the version (or other signaling mechanism) would not make sense. Thus the entity that is issuing the work needs to set the version as that entity should be receiving and validating new blocks and transactions, and can therefore make the decision about whether it is ready to enforce the new rules.
Does the vote only count if you win a block?
Yes. The only way for other nodes on the network to know what you are signaling is by receiving a valid block from you. So you have to produce valid blocks to be counted.
