Building Live Voice Agents with Google’s ADK

Imagine a morning commute where your app holds a real conversation with you.

I picture that commute often.
What if the agent could plan your day, research a topic, and read a multi-speaker podcast aloud?
Wouldn’t that change how we use voice interfaces?

What this course teaches

The new short course from DeepLearning.AI shows you how to build live voice agents with Google’s ADK.
You learn to process speech input, reason with large models, and respond in natural audio.
You learn modular ADK pieces like models, tools, memory, and orchestration.
You learn to add guardrails, persistent memory, and automated evaluations.
You learn production steps using Vertex AI Agent Engine.

Why it matters

Voice moves computing beyond screens.
Why is that important to you?
Because voice can make tools faster, hands-free, and more human.
Because agents that coordinate multiple specialists — planners, researchers, writers — are closer than you think.

Course format and instructors

The course is about 1 hour and 15 minutes long.
It includes 9 video lessons and 5 code examples.
Instructors are Lavi Nigam and Sita Lakshmi Sangameswaran, both Machine Learning Engineers at Google.
You can preview the course here: https://www.deeplearning.ai/short-courses/building-live-voice-agents-with-googles-adk/

Quick examples

Build a simple voice agent that answers questions.
Extend it with a research agent that drafts a podcast script.
Add Gemini text-to-speech to produce multi-speaker audio.
Think of it like assembling a crew for a short film.

My takeaway

I recently skimmed the lessons and tried a demo agent.
The primitives for session, state, and memory make complex flows manageable.
This is practical work you can apply today.

Looking ahead

Voice agents will become more capable and more present in daily life.
Will you build the next useful voice app?
Enroll while access is free during the beta.

Full course: https://www.deeplearning.ai/short-courses/building-live-voice-agents-with-googles-adk/