NVIDIA's breakthrough voice model. 170ms latency. Full-duplex. Ship today.
import personaplex
client = personaplex.Client(api_key="...")
session = client.create_session(
voice="NAT-F2",
persona="You are a helpful assistant"
)
async for response in session.stream(audio_input):
play(response.audio)NVIDIA's January 2026 release changes everything about voice AI.
Listens while it speaks, just like humans do
Fastest open-source voice model available
Barge-in support, no awkward pauses
Based on NVIDIA's open-source release
Natural conversations with support for interruptions, overlaps, and back-channel responses. No more awkward turn-taking.
~170ms response time for real-time interactions. Fast enough to feel like a natural human conversation.
Define AI personality through voice and text prompts. Create consistent characters for your voice applications.
Choose from 16 high-quality voices out of the box. Multiple accents, genders, and speaking styles included.
Pay only for what you use. No hidden fees, no long-term commitments.
| Provider | Price/min |
|---|---|
| OpenAI Realtime | $0.06-0.24 |
| Bland AI | $0.07-0.09 |
| PersonaPlex | $0.08 |
PersonaPlex powers the next generation of voice experiences.
Build AI phone agents that handle customer calls naturally, with real-time responses and smooth interruption handling.
Create engaging AI characters for games, entertainment, and companionship apps with consistent personalities.
Power real-time translation services and accessibility tools with ultra-low latency voice processing.
Get your API key and start building in minutes.
Stream audio to our API and receive responses in real-time.
Create voice agents, assistants, and conversational apps.