1
1
1 year of experience

Voices is one of the main physical senses of a human being. Throughout history, serenades have been known to help move hearts. Using artificial intelligence to extract biomarkers or characteristics in voice can help improve matches. There has been more studies recently of vocal analysis by artificial intelligence to help detect diseases (cardiovascular, neurological, etc.) and we believe that voice can be used to help improve relationship chemistry. For our project we looked to demonstrate from a prototype how we are training our model to best estimate what voices our user would "like." We used user inputs as well as shimmer, pitch, pitch variance, and harmonic to noise ratio to see if there were any patterns. In addition, we used Cohere's embeddings to help establish relationships. In our analysis, we saw that Cohere's embeddings clearly was able to detect and distinguish between genders where user inputs clearly preferred one gender over another. We also saw promising relationships in vocal characteristics such as pitch, shimmer, and harmonic noise ratio where further training could help enhance accuracy. Overall, the dating app is currently heavily weighed on appearance alone. Our app and project intends to change that. By studying further voice of existing couples/married relationships, we can further improve our model. There are also other biomarkers that are readily available that we can use in the future to improve model accuracy. Please see our video demo here: https://www.loom.com/share/8ac63653b3c84ced8b8836fbf9c3cf3b?sid=24315fa5-6a20-4422-aec1-4a813516b06d Please see our presentation here: https://docs.google.com/presentation/d/10vpRYgltNhEcz5yOhY6UK-9cK4PQkkBpQ5gudniXm3c/edit?usp=sharing
21 Jan 2024