Author: Louis Juste
Voicera Sincerity is a new AI-powered tool that evaluates verbal and non-verbal cues from video and audio communications. Chandra De Keyser, founder and CEO tells us more about the tool and its potential use cases for different sectors.
You just launched Sincerity, an Emotion AI platform that analyses people’s behaviours and sincerity. What inspired you to launch this platform?
The massification of LLM means that we can now analyze, summarize, and understand any verbal communication. Video calls transcripts, emails, chats, documents.
What remains to be truly understood is the non–verbal component of human communication: facial expressions, tone of voice, and body language. All of these have profound meanings that may be missed in massive amounts of video communications.
This is what drives us: bringing emotional intelligence to help people get the full picture: what they say and how they say it. Our Chief AI, former Chief Data Science of SAP, developed a family of AI models to provide actionable insights from omnichannel communications.
The most innovative AI model he and his team developed is Sincerity, a new frontier of AI, which we have just brought to the market.
What kind of data did you train your platform on?
The Sincerity AI model is trained on academic datasets along with synthetically generated data which are proprietary. The academic datasets are a comprehensive collection of multimodal data, including video and audio recordings with labelled examples of truthful and deceitful statements. Their size is substantial, with thousands of hours of recordings, ensuring a wide variety of scenarios and contexts.
Additionally, synthetic data is generated to supplement the dataset, providing even more diversity and helping the model to generalize effectively to new, unseen data.
“Our ambition is to be the reference company for AI that reinforces trust in human communications.”Chandra De Keyser, founder and CEO of Voicera
What are some of the use cases of Sincerity? Who is your main audience?
First of all, it is important to note that humans are not very good at detecting lies. Psychologists Bella DePaulo of the University of California, Santa Barbara and Charles Bond of Texas Christian University reviewed 206 studies involving 24,483 observers judging the veracity of 6,651 communications by 4,435 individuals. Neither law enforcement experts nor student volunteers were able to pick true from false statements better than 54% of the time – just slightly above chance. In individual experiments, accuracy ranged from 31% to 73%, with the smaller studies varying more widely.
We are exploring different market segments such as legal professionals, HR/recruitment, audit, internal investigations, insurance/finance fraud detection, Investment / Business negotiations, as well as B2C use cases in interpersonal relationships.
How accurate is Sincerity in distinguishing between truthful and deceiving behaviour or speech?
The model outputs probabilities for two classes: “Likely untrue” and “Likely true”, with a thresholding approach to introduce a third class, “Neutral”. Based on Leave-one-out cross-validation, the test set results indicate an average accuracy of approximately 78%.
What do you hope to achieve with Sincerity by the end of the year?
Our ambition is to be the reference company for AI that reinforces trust in human communications. We also want to achieve product-market fit in at least one market segment whose size is $1bn of annual revenue and enhancements of the Sincerity AI model.
Interview courtesy of our content partner Silicon Luxembourg
Comments