Discussions

Hi HeyGen Team,

I’m currently implementing a real-time integration where the HeyGen Interactive Avatar interacts dynamically with our backend and database using TaskType.REPEAT. This setup works great for backend-driven logic and data retrieval.

However, in Repeat Mode, the user currently needs to manually toggle the mic on and off for every input. I’d like to create a continuous, natural voice conversation — similar to TaskType.TALK — but still retain backend control over responses via TaskType.REPEAT.

My main question:
Can methods like startVoiceChat() or startListening() / stopListening() be used within TaskType.REPEAT mode to maintain an always-on mic and seamless interaction?

The SDK documentation mentions that voice chat is tightly coupled with the internal LLM/knowledge base and that “custom avatar speech input is not supported when voice chat is enabled”.
So, I’d like clarification on:

Whether startVoiceChat() can coexist with TaskType.REPEAT (for backend-generated responses).

If not, is there any workaround to keep continuous mic input active while still routing processing to our backend?

Alternatively, can we dynamically inject or update the avatar’s knowledge base in real time during an ongoing interactive session (so it reflects backend data without restarting the session)?

Any implementation guidance, code snippet, or confirmation from your engineering team would be greatly appreciated.

Best,
Sanket Bari