NovaSynth by Noveum
Test your voice agent on the callers you can’t stage.
2 followers
Test your voice agent on the callers you can’t stage.
2 followers
Simulate realistic callers at scale with custom personas, scenarios, interruptions, noise, accents, and network conditions. NovaSynth runs those calls against your voice agent, scores audio and transcripts across 30+ dimensions, and surfaces the failures and fixes that matter to your team.
Hi Product Hunt 👋
I’m Aditi, co-founder of Noveum AI.
Voice agents have guardrails, scripted tests, and observability. The caller is still a variable
Real callers don’t work that way.
- The Interrupter: Cuts you off mid-sentence, talks over the agent, then demands an immediate answer.
- The Second-Guesser: Agrees, then changes their mind twice and asks for a completely different option.
- The Hard-to-Understand Caller: Speaks softly, switches pace, uses an unfamiliar accent, and repeats themselves.
- The Distracted Ones: Calling from a crowded, noisy place with people talking, traffic, and constant interruptions around them.
- The Frustrated Caller: Starts impatient, gets increasingly annoyed with every extra question, and is close to hanging up.
- The Unstable Connection: Audio drops, words get clipped, pauses appear unexpectedly, and the caller repeats themselves.
And you can’t keep relying on a human tester to recreate every one of those situations.
Why NovaSynth Exists
A year ago, Shashank and I were working on voice agent evaluation when someone said:
“The happy paths are working fine. I pray the agent works well when a weird customer picks the call."
That question stayed with us. It became the starting point for NovaSynth, a pre-production testing platform built around simulated callers.
KEY FEATURES
→ Build the caller, not just the test: Create personas and scenarios around accent, mood, behavior, interruptions, noise, network conditions, and caller intent.
→ Mix and match personas and scenarios: Create different callers for different situations, then test any persona against any scenario as your agent evolves.
→ Run real voice interactions: Simulate calls through real audio and telephony paths, capturing voice-specific issues such as latency, interruptions, and audio quality.
→ Evaluate beyond the transcript: Score calls across 30+ audio and 100+ transcript scorers, plus the business KPIs your team cares about.
→ Validate fixes before shipping: Use NovaPilot to recommend fixes, then backtest and re-run them against your own calls and traces.
We’re here today, Shashank and I will be around all day to answer any of your questions or doubts.
Start a Free Trial: https://bit.ly/4zOcEXT
Book a Demo: https://bit.ly/4gtBXae
And I’m curious:
What’s one caller you wish you could test your voice agent against, but have never been able to recreate?