Introducing Realtime Voice Chat

Samuel Su
on October 20, 2024We're excited to roll out Realtime Voice Chat, a groundbreaking feature that lets your chatbot communicate just like a real person! With this update, users can engage in natural voice conversations, making interactions faster, more engaging, and more intuitive.
🚀 Available on Standard and Above Plans
🔧 How to Enable Realtime Voice Chat
Realtime Voice Chat is not enabled by default due to its high resource costs. To activate it:
1️⃣ Go to Settings > Realtime Audio.
2️⃣ Choose the voice avatar that appears in the chatbot.
3️⃣ Set a default voice (users can change this later).
4️⃣ Preview available voices before selecting.
🎙️ Realtime Voice Chat Features
- ✅ Audio Playback in Chat Logs
- By default, original voice interactions are stored and available in the Chat Logs Dashboard for review.
- Disabling this option will store only text transcripts (generated by OpenAI's Whisper model).
⚙️ Adjustable Voice Chat Parameters
Fine-tune voice interaction settings for a seamless experience:
- Threshold – Adjust voice activity detection sensitivity.
- Prefix Padding – Set how much audio before speech recognition is included.
- Silence Duration – Define when the chatbot considers speech to have ended.
- RAG Limit – Control how much context is retrieved per message to optimize costs.
Users can also change voices dynamically via the voice selection overlay. 🌍 Multi-language support is available!
💰 Cost & Pricing
Voice-based interactions are processed through OpenAI's realtime API, which is more resource-intensive than text-based queries.
📌 Cost Optimization Tip:
- By default, the chatbot retrieves 10 RAG text chunks (~3,000 tokens). Lowering this to 5 chunks reduces costs.
- gpt-4-realtime-min-preview is more cost-effective than gpt-4o-mini, averaging 1 credit per message.
- Messages without context retrieval cost ~10 credits, while those with context retrieval cost 30-40 credits.
📊 View Credit Usage:
You can check cost per message in the Chat Logs dashboard by hovering over the question mark icon.
📇 Lead Information Collection
With Realtime Voice Chat, lead collection is fully automated! 🎯
- The chatbot can extract and store user details like name, phone number, and email directly from conversations.
- This data is synced to your leads dashboard, webhooks, and CRM.
- The chatbot remembers lead information for future conversations, even after a page refresh.
💡 Customize Data Collection:
Modify the chatbot's base prompt to gather user details before providing further service.
🎉 Get Started Today!
1️⃣ Enable Realtime Voice Chat from the Settings page.
2️⃣ Fine-tune parameters for cost efficiency and accuracy.
3️⃣ Enjoy natural voice conversations with automated lead capture!
🚀 Experience a truly conversational AI with Realtime Voice Chat today!
Check the Realtime Voice Chat page for more details.