Can anybody with practical experience developing with Wit.ai comment on how accurate and consistent it works? Is there any new and better working software behind this, compared to the current breed of frankly abysmal voice recognition software (Siri, Nuance etc)?