I’m developing an AI-powered tool that listens to a patient’s voice in real time, detects subtle emotional cues, and immediately presents those insights to clinicians. The core need is to identify emotional patterns (stress, calm, anxiety, depression, etc.) from spoken input so care teams can intervene early and personalize treatment. Here is what I’m looking for you to build and hand over: • A low-latency pipeline that captures live audio, performs signal pre-processing, and feeds it to a machine-learning model, all fast enough to feel instantaneous during a consultation. • A polished analytics dashboard—web or desktop—that visualises detected emotions, confidence scores, and time-series trends, with the ability to export session summaries. • Model training code and documentation: clearly explain data requirements, feature extraction steps, and how to retrain or fine-tune the system for new languages or dialects. • Clean, well-commented source code (Python, TensorFlow/PyTorch, or another proven stack) plus setup scripts so hospital IT staff can deploy on-prem or in a secure cloud. • Brief guidance on optional integration hooks (API endpoints, websocket events) for future connection to existing EHR or telehealth platforms. Acceptance criteria 1. End-to-end processing latency under 300 ms on commodity hardware. 2. ≥ 90 % class-balanced accuracy on a blinded test set containing at least five common emotional states. 3. Dashboard runs in the latest Chrome and Edge without additional plugins. 4. All deliverables pass a code review and a live demo session conducted over screen share. If you have relevant signal-processing or speech-emotion-recognition experience, I’d love to see a short note on your approach and any sample links.