The rate at which medicine has adopted technology (some of my doctors' offices are still on paper charts) versus its acceleration in the consumer space, I would see this happening with consumer tech and probably more likely to be something continuous versus a single session. Like an EKG versus a holter monitor.
Amazon has a device that already does this in a rudimentary way:
Amazon's new $99 fitness band has a pair of microphones that listen to your voice and analyze it to assess your tone and mood.
www.popsci.com
I'm not saying I think it will happen in either space. Just that if it were to happen, I think it would be more likely in the consumer space.
What companies already know about human nature that they haven't released due to ethical or creepiness concerns must be astounding, just by thinking of what they have chosen to release. Thinking of Gmail's auto-complete for example or it knowing when you might want an e-mail to be secure vs not baed on what you write. It's already a bit uncanny. I am sure they are holding back on their suggestions as to not appear too omniscient. (Apple on the other hand . . . well I think they're playing all the cards they have when it comes to AI, and it's not much.) Facebook was doing experiments testing human behavior based on what they showed people and how they responded in subsequent posts, but they supposedly stopped due to ethics of it.
I'm basing my assumptions on what I have not seen happen in the medical space and what I have seen happen in the consumer space. I'm not saying everyone should run out and self-diagnose, but as an example, it was so much easier and more possible for me to catch a sustained SVT episode using a cheap but FDA-approved-algorithm EKG device I bought myself (AliveCor) (the rhythm strip of which was later read and confirmed by a cardiologist) versus the much more expensive and old fashioned cumbersome testing (an ER trip and Holter monitor) that never caught it (converted to sinus tach by time I got to ER and didn't happen during Holter monitor). The consumer tech sensors right now are more rudimentary, but the investments in the algorithms are wider. If it weren't for privacy concerns, I wouldn't be surprised if a company like Google could already be quite predictive of human behavior if it used constantly listening microphones.
Privacy is a concern, though, and is obviously different with mental health than heart health. There's an AI service (and there are many like this) called Woe-Bot that does CBT on your phone. It came out from psychologists at Stanford, but I think they are collecting data to improve the service, so I'm wary of giving it too much information. That's probably an area where traditional medicine services still have an edge--the willingness to disclose. On the other hand, and with regard to passive continuous collection, when I was on facebook (which I haven't been for years), people were really willing to put their business out there. It wouldn't be too difficult to imagine with that volume of data they could start making a lot of inferences that could relate to mental health.
It also seems like tech purpose built for medical utility is always more expensive (vs consumer tech, including consumer medical tech), I suppose due to the lower volume of sales.