How much control should you have over the way a voice agent sounds?

A voice agent isn’t just about what it says. It’s also about how it says it, how quickly it responds, and whether it can adapt to the flow of a conversation.

We’ve been working on making voice behaviour more configurable at Vaani AI. Here’s what you can tune:

• Speaking speed: Adjust the pace of the agent’s voice depending on the TTS provider.

• Response timing: Control how quickly the agent responds after the caller stops speaking, and experiment with speculative response generation for faster turn-taking.

• Emotional expression: Configure supported emotions and voice tags so the agent can sound excited, calm, or more expressive where appropriate.

• Voice consistency: Tune parameters like stability and style to balance a consistent voice with more natural expression.

• Volume control: Callers can ask the agent to speak louder, quieter, mute, or restore the volume during a call.

• Natural audio details: Add optional breathing sounds and background ambience to make conversations feel less robotic.

• Entity handling: Give the agent more patience when callers are spelling out names, phone numbers, or other details.

One thing worth mentioning: these controls depend on the TTS provider. The same configuration won’t necessarily produce the same behaviour across different voices.

We’re interested in where people draw the line between a highly controlled voice and one that sounds more expressive.

When building voice agents, which matters more to you: consistent and predictable speech, or a voice that can adapt its pace, emotion, and delivery to the conversation?

If you’re experimenting with voice behaviour, check out Vaani AI @Vaani Developer Platform and try tuning these settings on your own agents. We’d love to hear what combinations work best in real calls.