Datasets:

on1onmangoes

/

GVMS042424_RT_AL_TEST

like 0

Hello, good morning. This is Amit Lambus speaking from New York, New York.

We are conducting a new demo for the clients. Let's see. Let's say who our clients are.

Let's say it's Ara of Lee Wang and Fatima. I've decided to come in for this meeting with Dr. Diego.

I might be a technician sitting in on this group session or I could be something ambient that the doctor could do themselves.

So this gives you an ability to kind of be ambient as well. The purpose of this meeting, let's say, is for just a quick evaluation of the current situation because the purpose will drive a lot of the downstream aspects of this.

As you can see, most of this application is focused on the vocal voice or the acoustics and the voice piece and speech. Patterns of the patient, as well as the doctor to diagnose, and we hate the word diagnose because diagnosis is not destiny, but looking at the symptoms and the symptoms are not just dictated by the language, but also the diction, the pace, you know, the pitch, the prosciadity, the number of pauses, and the filler words being used, and we've trained this in a multimodal way.

So at its peak, some of the models have about more than 2 million hours of data that trained it, and then we kind of fine-tuned it to find the needle in the haystack, then we showed it all the needles.

and then we kind of fine tuned it to find the needle in the haystack then we showed it all the needles and by that i mean expertly labeled data and construction of this data by a leading psychiatrist in new york itself who partnered with us in the initial stages of this as you can see we've kind of kept it fairly simple we don't believe in false precision but you can easily then update as we go along what the, you know, symptoms that the doctor himself that feels is.

And this is important for twofold. One is to check for biases both ways, the AI bias as well as the human bias and make sure we are actually individualizing it for the patient as their baseline.

So one size doesn't fit all, but it's actually mass produced as well. And we can take these best practices to everybody out there.

Once you're done, we can