How much processing power would you need to monitor every voice call in e.g. the US and apply voice recognition to every call?

Do things in series 24 really exist?