Mental health services are stretched thin. Counselor shortages, language barriers, and thesimple geography of underserved regions combine to leave the majority of people who needcare without access to it. Text-based digital tools have helped at the margins, but they misssomething fundamental: the voice carries emotional weight that no chatbot transcript canreplicate. This paper describes the design and evaluation of the Emotionally Responsive VoiceCounseling (ERVC) system, a web-based platform built around a voice-language foundationmodel capable of near-zero-latency interaction, emotionally modulated speech output, andsupport for over 100 languages.We used a mixed-methods approach that combined a systematic literature review, designscience research methodology, and structured scenario evaluation across ten clinical cases.Prototype testing showed clinically acceptable therapeutic alliance scores (mean WAI-SF5.47/7.0), consistent empathy ratings, and full adherence to safety protocols across allscenarios. Five use cases drove the design: round-the-clock crisis support, multilingualcounseling access, structured skill practice, passive mood monitoring through vocalbiomarkers, and sustained therapeutic companionship between human therapy sessions. Thiswork makes the case that emotionally intelligent voice AI, built on a principled ethicalframework, can meaningfully extend the reach of mental health care.
Sripooja Mallam (Tue,) studied this question.