Voice Activity Detection (VAD) and noise handling are critical components that determine how natural and responsive your voice conversations feel. Ultravox provides an intelligent, multi-layered approach to managing audio input that works seamlessly out of the box.

Our Philosophy: Smart Defaults

At its core, Ultravox handles VAD complexity for you. Our system uses multiple models working together to provide a great experience out of the box.

We are very aggressive about triggering VAD by default because:

  1. Latency Matters → Users expect immediate responses in voice conversations.
  2. Smart Models Compensate → Neural VAD and noise cancellation handle edge cases (e.g. false positives, interruptions).
  3. Context is Key → The conversation state helps determine when users are truly done speaking.

Additionally, Ultravox Realtime provides vadSettings that can be customized if and when needed. However, customizing VAD involves a series of tradeoffs.

Key Benefits

Low Latency → Aggressive VAD settings combined with intelligent models minimize response delays.

Natural Conversations → Context-aware models understand conversation flow and turn-taking.

Robust Audio Handling → Built-in noise cancellation and background speaker management.

Minimal Configuration → Works excellently with default settings for most use cases.

When to Customize

The default VAD settings are carefully tuned and work well for the vast majority of applications. You may need to adjust VAD settings for extremely noisy environments, accessibility needs, or other specialized situations.

If you are experiencing issues:

  1. Start with troubleshooting → Verify your audio setup and environment.
  2. Test with default settings → Ensure the issue isn’t environmental or hardware-related.
  3. Make incremental changes → Adjust one parameter at a time.
  4. Monitor the trade-offs → Test thoroughly with real users and use cases.

The system is designed to be intelligent and adaptive. Trust the defaults unless you have a specific, tested need to change them.

Proceed with Caution

Changing VAD parameters is a trade-off game. Every adjustment has consequences for latency, accuracy, or user experience. We recommend extensive testing when modifying default settings.

What’s in This Section