Longer description of your proposed project

The proposal involves funding research at UC Berkeley, conducted by a group of students, visiting researchers, and professors. The goal of the project is to conduct research on voice biomarkers; quantifying speed, inflection, intonation, mood, memory, etc. to make precise estimates about mental health, namely the likelihood of neurodegeneration, and progress of mental well-being. The central aim of the research is to conduct research into the plausibility of using voice biomarkers as screening methods and early prevention mechanisms for Alzheimer’s, Parkinson’s, Depression, PTSD, and more. The first aim is to understand the correlations between voice and the development of neurodegenerative disease through computational methods. The next step is to generalize patterns in the population to gain a high accuracy in predicting disease. Finally, we want to create an accessible model that allows any person to use the technology through mobile software.

The project involves conducting its own research using AI and machine learning models as well as digitizing existing data from initial findings recently published in clinical studies and academic papers. Our initial work has focused on developing a classification algorithm for dementia prediction based on text-based input with an accuracy score of 96%, and we are currently working to make it voice-based. Our current research focuses on encoding speech properties such as loudness, tone, pitch, fluency, articulation, pauses, and context from arbitrary scales into vectorial data. We will then use machine learning models to conduct classification using past clinical trials as a source of training data. Finally, we will maximize model accuracy and efficiency to correctly classify unique voice data, to predict the likelihood of various neurodegenerative diseases. Our pilot research will focus on stress, mood, depression, and dementia with the hope of extending our research to more neurodegenerative diseases in the future as more clinical trials become accessible.

In the long term, we envision an application of this research launched into a usable, personalized, scalable API. The voice biomarker software will be launched alongside an interactive software featuring a talking voice bot, such that people can talk to a virtual companion, freely and in an unstructured environment, while allowing for their data to be screened. This allows for an easy-to-use program remotely at home, without the need for a hospital environment. The talking virtual companion is a feature of our API that is already working; it can currently conduct natural, interactive, and entertaining conversations while remembering the user’s previous dialogue, allowing it to track memory progress and its respective degradation.

With this research, a future of digital and advanced early detection screening is possible, allowing for decentralized, affordable, and accessible healthcare. With the new state-of-the-art possibilities of biomarkers, the general ambition is to detect disease up to 10 years before its onset. Such research has the power to transform mental health treatment. According to the UN, up to 1 billion people, nearly one in six of the world’s population suffer from neurological disorders, from Alzheimer’s and Parkinson’s diseases to strokes, multiple sclerosis, and infections. The large cognitive decline of the population is costing national healthcare billions in budget, for an increasingly costly healthcare system to the individual American. With the power to predict, detect, and manage certain neurological illnesses, we have the power to change millions of lives.

Describe why you think you're qualified to work on this

My name is Amelia Lubelska. I am an international student from Warsaw, Poland, studying at UC Berkeley. I am pursuing an undergraduate degree in Molecular and Cellular Biology with a Specialization in Neurobiology. I am simultaneously pursuing Data Science and a certificate in Entrepreneurship from the Sutarja Engineering and Entrepreneurship Center at Berkeley. I am passionate about the intersection of neuroscience research, healthcare solutions, and personalized therapy. I am currently conducting research at the Computational Cognitive Neuroscience Lab at Berkeley, studying reinforcement learning and reward mechanisms in humans. I am also the Lead of the Education Division in the organization Neurotech@Berkeley, as part of which I teach a class at Berkeley titled Introduction to Neurotechnology, sharing my passion for neuroscience with over 100 students. I share my accomplishments to hopefully demonstrate that despite a young age, through my research and involvement, a passion for understanding the idiosyncrasies of the brain evolves, a peculiar trait of the brain whose applications can be lifesaving if ultimitely understood.

I have experience both in wet lab as well as computational research, I frequently give lectures and speak publicly, I am a project manager to over 30 students at university, and above all, I cherish the ability to engage in discussion with them to create novel moonshot ideas. This is how voice biomarkers for mental health started: just one conversation with two visiting researchers at UC Berkeley prompted the start of a research team. I have been working with Lorenz and Alessandro every day ever since, expanding our team, conducting research, and building AI models.

Lorenz Pichler, from Austria, pursued his master’s in data science at IE University in Spain before becoming a visiting researcher at UC Berkeley. Passionate about natural language processing (NLP), he focuses on speech and multimodal classifiers. Before delving into academia, Lorenz showcased resilience and discipline as a professional sailing athlete, proudly representing the Austrian national team. With a tech consulting background and three years of valuable experience, he combines practical insights with theoretical knowledge. Inspired by a personal connection to the challenges of cognitive health, particularly witnessing his grandmother's struggle with dementia, Lorenz is deeply committed to improving diagnostic methods. In his pursuit, he strives to translate academic research into an accessible and cost-effective solution for the healthcare industry.

Alessandro Neri completed a bachelor’s degree in business administration from Bocconi University in Italy, followed by a master's degree in data science from IE University in Spain. For over a year, he has been focusing on voice biomarkers and generative AI. Currently, he works as a pioneering researcher at UC Berkeley, where he seeks out and exploits business opportunities in the field of AI. Together with Lorenz, Alessandro previously worked on real-time automatic speech recognition models, integrating them into AI frameworks to enable ultra-realistic, real-time voice conversations.

When we first met, we began exploring the incredible possibilities of this technology and contemplated how revolutionary it could be to use it for diagnosing cognitive abilities. This application could enable the widespread and affordable diagnosis of millions of people who previously lacked access to such services. Since we had this conversation three months ago, we have been working every day together. Together, we make a formidable team, fueled by our shared passions, and supported by a range of skills from AI, Neuroscience, and Entrepreneurship. Behind our team, we have the support of the Saturja Engineering and Entrepreneurship Center, along with fellow students, researchers, and professors at UC Berkeley and IE University, who are collaborating with us to bring the best minds together on a unique and intricate problem. With the grant, we are certainly qualified to tackle this moonshot task that has the potential to revolutionize modern neurological healthcare.

Other ways I can learn about you

www.linkedin.com/in/amelia-lubelska (sorry, no twitter, no ACX)

How much money do you need?

70-100k. Primary goals of funding will be to support the core team of researchers’ salaries. Next, funding will go to expanding the team and cross-university collaborations to increase rapidity of research. Finally, funding will be necessary to acquire data from non-open clinical trials and research. If our research is as promising as expected, we will be looking at patenting and FDA paths in the future.

Links to any supporting documents or information

No response.

Estimate your probability of succeeding if you get the amount of money you asked for

While the research itself feels like a moonshot, there is substantial evidence in first clinical trials and research that we are capable of building a voice biomarker prediction model for neurodegenerative illnesses. If we receive the necessary funding, the team will work at 100% capacity and efficiency. From there, the likelihood of success will depend on how this solution compares to existing screening methods in accuracy. Given neuroscience, AI, and startup market research, we have reason to believe that our science is solid with 75% estimation of success. Additionally, we hope to have conclusive results by 2025, provided by accelerated growth given appropriate funding.