Let's Talk Soul Food

Let's Talk Soul Food is a culturally inclusive, voice-first AI agent that conducts dietary and health profile interviews for the African American community. Powered by Speechmatics' Global English model, it accurately understands AAVE and Southern regional accents, ensuring spoken health data is captured reliably the first time. THE PROBLEM Voice-powered health apps have long failed African American users in two ways. First, traditional STT engines struggle with AAVE and regional accents, misinterpreting natural speech patterns. Second, when an STT engine mistranscribes a medical condition or dietary restriction, the downstream AI is working with corrupted data. An AI dietician cannot build a safe profile if the voice intake layer fundamentally misunderstands the user. WHY SPEECHMATICS We built on Speechmatics' Batch API with the Enhanced operating point for its superior performance on diverse conversational speech. This ensures our 7-question dietary intake loop captures culturally specific terms and regional pronunciations without introducing phonetic errors, giving our LLM the clean data it needs to generate personalized, healthy soul food recipes. TECHNICAL HIGHLIGHTS We tuned our polling loop to a 0.5-second interval, achieving a responsive 2-second conversational beat without triggering rate-limit penalties. We benchmarked Speechmatics TTS and confirmed near-zero leading silence under 85ms, eliminating dead air between turns. We implemented a Phonetic Forgiveness instruction in our LLM prompt, allowing the AI to gracefully handle minor transcription variations and correctly interpret the user's spoken intent. THE IMPACT Let's Talk Soul Food proves that Speechmatics' acoustic modeling can make AI health tools accessible, personalized, and culturally authentic for communities historically underserved by voice technology. By getting the speech layer right, every user receives the same quality of personalized health guidance regardless of how they speak.