Effective manager‑employee communication is essential for retaining top talent and improving low performers, yet traditional training is expensive. Text‑based chatbots scale well but cannot provide realistic spoken rehearsal, which is needed for managers to build confidence. This paper introduces Conversation Coach, a voice‑first AI system that lets managers rehearse challenging workplace dialogues in a natural spoken setting. The system tackles three challenges: delivering low‑latency interactions with strong language understanding; offering adaptive conversations through configurable bot personalities that mimic different employee types; and generating personalized feedback on content and policy compliance. We built both an end‑to‑end speech‑to‑speech model and a cascaded pipeline that combines automatic speech recognition, a large language model, and text‑to‑speech synthesis. The end‑to‑end approach reduces median latency by threefold, supports native barge‑in, and cuts estimated cost by eightfold, while the cascaded approach provides superior reasoning for higher coaching quality. We deployed the cascaded architecture in production; over six months more than 40,000 managers used it, primarily for the most difficult conversations.
Review