Doing natural language processing is still a very difficult task, especially in situations where you don't first "train" the computer to recognize your voice, and/or where there's a noisy background.
We can't yet do untrained language recognition in a TV or cell phone. So when you use Siri or Cortana or a voice-operated TV, your voice is recorded and uploaded to a huge server farm for analysis and processing, and the results are sent back down to the device. That's why Siri won't work if you don't have an Internet connection, and it's why this TV uploads everything you say to company servers.
