Skip to main content

Voice Responses

Nova can now speak her responses! When you use voice input, Nova will automatically respond with voice in addition to text.

How It Works​

  1. Tap to Record: Tap the microphone button and speak your question
  2. Automatic Model Switching: If using a non-voice model, automatically switches to GPT-4O Realtime
  3. Automatic Transcription: Your speech is transcribed in real-time
  4. Voice Response: Nova responds with both text AND voice
  5. Smart Truncation: Long responses are intelligently shortened for voice

Model Compatibility​

The microphone button is always visible, regardless of your selected model:

  • Voice-Enabled Models: GPT-4O Realtime, GPT-4O Realtime Preview
  • Other Models: Clicking the mic automatically switches to a voice-enabled model
  • Seamless Experience: No need to manually change models for voice

Smart Response Truncation​

Nova intelligently adapts her voice responses based on content type:

Standard Responses (< 75 words)​

  • Spoken in full
  • Natural, conversational tone

Longer Responses (> 75 words)​

  • First paragraph or 75 words spoken
  • Adds "I've written more details below"
  • Full response available in text

Lists and Enumerations​

  • Announces total count: "I found 12 items"
  • Speaks first few items
  • Adds "I've listed all items in the text below"

Code Blocks​

  • Says "I've written some code for you"
  • Adds "Please review it in the text display"
  • No code is spoken (avoids confusion)

Voice Controls​

During Recording​

  • Stop Button: End recording manually
  • Auto-stop: Coming soon - silence detection

During Playback​

  • Volume Control: Adjust voice volume
  • Stop Speaking: Cancel voice playback
  • Visual Indicators: See when Nova is speaking

Technical Details​

Supported Models​

  • GPT-4O Realtime Preview
  • GPT-4O Realtime (when available)

Voice Options​

  • Shimmer (default)
  • Echo
  • Alloy
  • Nova
  • Fable

Audio Format​

  • PCM16 encoding
  • Real-time streaming
  • Low latency response

Best Practices​

  1. Clear Speech: Speak clearly for best transcription
  2. Natural Language: Talk as you would to a person
  3. Context: Provide context in your questions
  4. Interruption: You can interrupt Nova's voice response

Coming Soon​

  • Auto-stop after silence
  • Voice mode preferences
  • Custom voice settings per agent
  • Background noise reduction
  • Multi-language support