For loud calls to be stable in large-population evolutionary dynamics, the fitness benefit from the call (e.g., mate attraction) must exceed the fitness costs from the call (e.g., predator attraction).
Individual species can 'solve for' benefit > costs by any number of devious tricks. One classic trick used by noisy insects (e.g., cicadas) is 'predator saturation'. Imagine you're a predator that hunts noisy insects by following their call - any time you find one, you'll eat it - unless you're already so full that you literally can't fit any more insects down your gullet. If the noisy insects have a short breeding season, and are noisy for the whole of that season, then during the season you will very quickly fill up and be unable to eat, but you will only have eaten a tiny fraction of the insect population.
If the predators are already full, the costs of predation risk are very low. The benefits of mate attraction remain as high as ever. In that case, an insect's best strategy is to sing as loud as possible.