Cross-modal prediction in speech depends on prior linguistic experience