In a year or two, as voice devices become ubiquitous, consumer squeamishness towards having microphones implanted throughout a home will likely also decline. This may open up the possibility for constant streaming of audio to cloud for processing. With this, the issues around endpoint detection could be mitigated.
When using continuous speech recognition, the challenge becomes analyzing context to be able to determine the end of a command. In lieu of trigger, it would be understanding the context to know when the service was being commanded.
With this type of implementation, intelligent assistants can then have an additional element of their personality — their predilection for barging into conversations.
“Uhmm… actually you’re both wrong.”