Currently, voice assistants respond when explicitly invoked by a specific wake-word---such as ``Alexa,'' ``Siri,'' or ``Ok Google''---and serve common functions such as answering questions and controlling other connected devices. However, prominent players in the industry are trying to make voice assistants even more seamless. For instance, Google Home and Alexa offer a continued conversation feature (or ``follow-up mode'') to allow users to make follow-up requests after the first command without repeating the wake-word. Alexa also has a feature called Drop In, which allows whitelisted users to begin an audio or video call without the other party manually picking up the call, effectively allowing remote parties to listen at any time. Looking forward, Amazon and Google have patented the ability of voice assistants to automatically extract keywords from ambient speech and use that to provide targeted ads to users. With advancements in natural language processing technology, we can therefore expect that the next generation of voice assistants will offer proactive assistance based on the audio signals and conversations acquired from the acoustic environment. We refer to such devices as ``always listening voice assistants'' and explore expectations around their potential use. We have performed a178-participant survey investigating the potential services people anticipate from such a device and how they feel about sharing their data for these purposes. We investigated the following research questions:

​

RQ1: What services do users expect always-listening voice assistants to provide based on their conversations?
 

RQ2: How useful are these potential services? Do people feel comfortable with them?

RQ3: How do factors such as demographics, perceived comfort, and service usefulness affect users' intentions to allow the voice assistant to access their conversations in exchange for services?