project1 min read
voxora: full-duplex voice AI with sub-second latency
Autonomous phone and web voice receptionist for service businesses. Full-duplex 24kHz linear PCM streaming over WebSockets with AssemblyAI voice agents, DTMF synthesis, and real-time audio visualization.
Traditional IVR phone systems are awful: "Press 1 for hours, press 2 for billing, wait on hold for 14 minutes." Most callers just mash zero until a person answers.
Small trade and service shops miss calls every day while they're on a job or driving between sites. Those missed calls are lost revenue.
Voxora is a voice receptionist that answers instantly, sounds natural, understands context, and can actually take actions.
What it does
A caller dials in or starts talking on the web. Voxora listens in full duplex:
- Sub-second conversational latency: It doesn't wait for awkward silence timeouts before replying.
- Natural interruptions: If the caller interrupts mid-sentence, Voxora stops speaking immediately and processes the new turn.
- Function calling in active calls: Books appointments, checks service availability, captures callback numbers, and triggers notifications.
- Browser voice console: Real-time 24kHz audio capture via the Web Audio API, DTMF synthesis, and an audio-reactive visualizer.
Stack
Next.js 16 frontend and API routes. High-throughput bidirectional WebSockets streaming raw 24kHz linear PCM audio directly into AssemblyAI's voice agent infrastructure. Tool execution runs async with fallbacks when connectivity wavers.