Everyone's shipping voice agents this year. Receptionists, collections bots, lead qualifiers, support lines. The demos look great.

Then they hit a real phone call.

I lead growth at a voice AI company, so I sit in on a lot of these conversations. And the thing that surprised me is how rarely the LLM is the problem. It's almost always the voice layer underneath it. The latency that was 90ms in the demo, suddenly climbs to 400ms on a real phone call. The voice that sounds great in English and falls apart the moment someone switches to another language mid-sentence. The TTS that reads a pincode like a phone number. Or the per-minute bill that was fine at 10,000 minutes and became the biggest line item at a million.

Real-time voice AI has always forced a trade-off: fast, natural, or cheap pick two. Murf AI's Falcon 2 to kills that trade-off entirely.
What it does

Falcon 2 is a real-time text-to-speech model purpose-built for voice agents receptionists, call centers, support lines, conversational AI running at $0.01/minute, about a fifth the cost of comparable models.