Hear and See Emotion with inVibe: Representative Patient and HCP Listening Platform

Summary

Have you ever wanted your insights to go beyond just what people say and understand how they feel when they say it? Speech emotion recognition (SER) uses AI to analyze tone, pitch, and intensity to give us deeper understanding of patient experience, doctor interactions, and market trends.

Check out the video below to hear inVibe’s Christopher Farina explain how this technology is transforming qualitative analysis and leading to more meaningful insights.

Transcript

My name is Christopher, and today we'll be talking about how we at inVibe use machine learning and AI to measure emotion. To begin, let's look at how computers measure emotion. Now, speech-emotion recognition is an application of AI that uses machine learning models to measure and classify emotions as conveyed in the voice itself, not in the words spoken. Our model does this by taking into account vocal qualities that humans can easily distinguish, like pitch, or loudness, intonation, the rise and fall, pace, speech rate, fluency, your smoothness, and a bunch of others that anyone would really struggle to articulate, even with training. These vocal qualities then get laddered up to three acoustic measures that are much more interpretable. Activation, which is the interest or enthusiasm in the voice. Valence, the emotional positivity. Dominance, the level of control or assuredness. And we use these measures, not just because they're more interpretable, but because they've been shown to be more robust, more adaptable, and more accurate compared to measuring emotions directly. Now, let's go through an example of how SER can uncover insights that's not readily available from just the language used. In this particular case, we asked oncologists to share their reactions to clinical trial data released at a conference. The first ONC we’ll hear is reacting to an abstract that covers the results of a phase one study of a novel molecule. The second one is reacting to a phase two study of a novel combination but involving two familiar molecules. As you listen, keep an eye on the circular chart on the right called the circumplex that plots the level of enthusiasm on the y-axis and the level of positivity on the x-axis in the voice throughout their response.

“The Savart study is very, uh, kind of narrow and interesting study. Particularly when looking at patients with both. Both the carious mutation and with intracranial disease. So it's a very narrow kind of, uh, group of patients, uhm, you know, for those patients. The fact that there is some signal of efficacy is interesting to say the least.”

“Uhm, this is a very interesting combination, uh, to, uh, in treating, uh, lung cancer. I think mechanistic-wise, this is a brilliant combination. And the data also looks very good.”

Both of these responses cover many of the same topics in much the same language. However, I’m sure you heard and saw in the circumplexes some differences between the two ONCs. Discussion of abstract one clustered in that lower left quadrant, and discussion of abstract two in the lower right.

Now, generalized to the broader sample, these differences reveal that ONCs are more positive about abstract two, that’s further to the right, which indicates a greater willingness to adopt this treatment once it’s FDA-approved. Good news for the sponsors of abstract two, and a vital warning for those sponsoring abstract one.

Thank you for watching our brief video! We hope we’ve given you a glimpse into how novel technologies can be leveraged to gain more insight from language data.