Skip to main content
Configure how your agent processes and understands speech.

​
Transcript Model

Select the speech-to-text engine:
  • Deepgram [Nova 3 General]: Best accuracy for general conversations
  • Deepgram [Nova 2]: Good balance of speed and accuracy
  • Whisper: Open-source option
Recommended: Deepgram Nova 3 for production use

​
Transcript Language

Select the primary language for transcription:
  • English (India)
  • English (US)
  • Hindi
  • Multiple other languages supported
Important: Match this to your target audience’s language and accent.

​
Custom Keywords

Add domain-specific terms for better recognition: 
e.g., Acme, Project Zeta, phonefix, CRM
Use Cases:
  • Company names
  • Product names
  • Technical terms
  • Industry jargon

​
Smart Format

Automatically formats numbers, dates, and currency in transcripts.
  • Enabled: “two thousand twenty five” → “2025”
  • Disabled: Raw transcription
Recommended: Enable for better readability

​
Numerals

Converts spoken numbers to digits in transcripts.
  • Enabled: “three hundred” → “300”
  • Disabled: “three hundred”

​
Allow Interruptions Overall

Enables customers to interrupt the agent while speaking.
  • Enabled: Natural, human-like conversations
  • Disabled: Agent completes sentences before listening
Recommended: Enable for better user experience

​
Allow Interruptions Initially

Controls interruptions during the greeting message.
  • Enabled: Customer can interrupt greeting
  • Disabled: Greeting plays completely
Recommended: Disable to ensure greeting is heard

​
Interruption Sensitivity

Controls how easily customers can interrupt (0-4 scale). Settings:
  • 0: Very difficult to interrupt (agent rarely stops)
  • 1: Difficult to interrupt
  • 2: Moderate (recommended for most cases)
  • 3: Easy to interrupt
  • 4: Very easy to interrupt (agent stops frequently)
Recommended: 2 for balanced conversations