VoxFlow — AI Vocal Coach

VoxFlow helps singers practice with useful guidance even when a vocal teacher is not available. Lyra, a conversational coach powered by the AssemblyAI Voice Agent API, leads each session through natural speech. The user can choose an exercise, request technique guidance, start or stop accompaniment, transpose a scale, change playback speed or volume, and finish the session without touching the interface. The browser measures the user’s vocal signal locally instead of asking the language model to infer performance from a transcript. Diaphragmatic Breathing records completed cycles. Lip Trill Scale evaluates pitch, stability, and octave alignment. Vocal Sirens measures range, continuity, direction, and smoothness. Notes Practice asks the user to hold four sequential target notes within ±30 cents for 500 milliseconds and shows live tuning feedback. When the user ends the session, VoxFlow turns these measurements into a bounded performance summary. Lyra speaks feedback grounded in those results while the interface displays strengths, improvement areas, and a concrete next action. The final report can be downloaded for later review. VoxFlow combines AssemblyAI’s low-latency conversation, turn taking, speech output, and JSON Schema tool calling with deterministic browser-side audio analysis. This separation keeps the coaching natural while ensuring that performance claims come from measured evidence.