Kintsugi Shuts Down, Open-Sources Voice Biomarker AI for Depression and Anxiety Detection
Kintsugi, a startup developing AI to detect depression and anxiety from voice, is shutting down and open-sourcing its technology due to costly FDA regulatory hurdles. The company’s models analyzed brief speech samples and showed significantly higher depression detection rates than standard questionnaires in a health payer trial. The release aims to empower global research into voice biomarkers for mental healthcare.
Kintsugi, a startup that developed AI voice biomarker technology for detecting depression and anxiety, is shutting down its commercial operations and releasing its research and technology into the public domain. The company's AI models, which analyze acoustic signals in speech to identify clinical markers of distress, faced high costs and lengthy timelines from the U.S. Food and Drug Administration's regulatory process, making the venture-backed business model unsustainable.
Voice has emerged as a promising biomarker for health monitoring. Scientists are using artificial intelligence to analyze subtle shifts in pitch stability, vocal clarity, and acoustic noise that the human ear would likely miss. These changes can point to conditions such as benign vocal cord nodules, polyps, or even early indicators of laryngeal cancer. AI may flag when something is off much earlier than typically caught, potentially serving as a first-pass screening tool that prompts clinical exams sooner. This approach is part of a broader shift toward passive, always-on health monitoring, where devices like smartphones and smart speakers capture voice data and AI learns an individual’s normal patterns to flag deviations.
Kintsugi’s platform used novel machine learning and deep learning to attribute clinical depression and anxiety from just 20 seconds of free-form speech. The models are language-agnostic and protect privacy by analyzing how patients speak rather than what they say. In one deployment with a major health payer, standard patient health questionnaires indicated only 3% of members had depression, but Kintsugi’s voice biomarker AI detected moderate-to-high depression in 33% and severe depression in 14% of recently discharged emergency department and maternal health patients.
Despite these results and commercial relationships with health providers and payers, the cost of FDA De Novo clearance proved too great for the venture-funded startup. The company had raised $8 million in seed funding in 2021 and $20 million in Series A funding in 2022 before ultimately deciding to release the technology open source to allow a global community of scientists to advance voice-based identification, triage, and monitoring without proprietary barriers.