AI Voice & VoIP

Introduction

AI Voice & VoIP enable your AI agent to interact with users through natural, human-like voice conversations.

With voice enabled, your agent can understand spoken input, respond using speech, and handle conversations across voice channels such as website voice widgets and inbound toll-free phone calls. This is ideal for customer support, sales inquiries, and high-touch interactions where speaking feels more natural than typing.

What AI Voice Enables

With AI Voice enabled, your agent can:

• Understand spoken user queries

• Respond with natural voice output

• Handle real-time voice conversations

• Support accessibility and hands-free interactions

• Offer a more human customer experience

Voice works alongside chat and follows the same training, persona, and behavior rules.

Where to Configure Voice

All voice configuration is managed at the agent level.

To access voice settings:

1. Open your AI agent from the My Agents page

2. Go to Channels

3. Select Voice

This section controls all voice-related behavior.

How VoIP Works

The VoIP flow is simple:

1. A customer dials your toll-free number

2. The call is routed via Vonage or Knowlarity

3. Your AI Voice Agent answers the call

4. The agent understands and responds using voice

5. The full call is logged in Exei

This enables AI-powered phone support without manual call handling.

Voice Settings

Voice behavior is configured from Channels → Voice → Settings.

These settings apply to both:

• Website Voice SDK

• VoIP (toll-free calls)

Language

Choose the language your agent should speak and understand.

Selecting the correct language improves:

• Speech recognition accuracy

• Response quality

• Overall conversation flow

Voice Accent

Accent controls how the spoken voice sounds.

You can:

• Choose from available accents

• Preview the accent before saving

This helps align the voice with your audience or region.

Text-to-Speech (TTS)

TTS controls how written responses are converted into speech.

Different models may affect:

• Naturalness

• Speed

• Clarity

Speech-to-Text (STT)

STT controls how spoken input is converted into text.

Accurate STT ensures:

• Better intent understanding

• Fewer transcription errors

• Smoother conversations

Greeting Message

You can configure a greeting message for voice interactions.

This is the first message users hear when:

• A voice session starts on the website

• A toll-free call is answered

Example:

“Hello, this is the AI assistant. How can I help you today?” 

Testing Voice Conversations

Voice testing is available directly in the Playground and is supported on all plans.

To test voice:

1. Open Test AI Agent

2. Switch to Voice mode

3. Speak to the agent and listen to responses

Testing helps validate language, accent, clarity, and flow before going live.

Voice & VoIP Conversations in Analytics

Voice and VoIP conversations are treated like all other conversations.

They:

• Appear in the Conversations module

• Include full transcripts

• Show intent, sentiment, and analytics

• Display fallback labels when applicable

This allows you to review and improve voice interactions just like chat conversations.

When to Use AI Voice or VoIP

Use AI Voice (Website Voice SDK) when:

• Users prefer speaking over typing

• Accessibility is important

• You want hands-free interactions on your website

Use VoIP (Toll-Free Calling) when:

• Customers expect phone support

• You want AI-powered inbound call handling

• You need 24/7 phone availability

• You want to reduce human call volume

Both can be used together for a complete voice experience.

Best Practices for Voice Setup

• Choose the correct language and accent

• Keep greeting messages short and clear

• Use conversational phrasing in responses

• Test voice regularly in the Playground

• Monitor fallback responses in Conversations

Voice experiences benefit from natural, simple language.

Common Mistakes to Avoid

• Not testing voice after changing settings

• Using long or complex sentences in voice responses

• Selecting the wrong language or accent

• Ignoring fallback signals in voice conversations

Regular testing helps avoid poor user experiences.

What’s Next?

Once voice is configured, you can deploy it live and monitor real conversations.

 Next User Guide: Conversations & Human Takeover