▶ Watch ↗AI Engineer World's Fair 20251:09:41
Full Workshop: Realtime Voice AI — Mark Backman, Daily
Read the full talk →Key ideas
Scroll to read ↓A voice bot needs more than speech generation: this workshop builds the audio pipeline, then explores interruptions, context, tools, testing, and selective routing between agents.
- A voice bot needs a working connection0:30 ↗
- Follow the audio through the processors3:29 ↗
- Read the sample bot from configuration to first turn8:42 ↗
- Choose the transport, then detect the user's turn16:14 ↗
- Connect phones and distinguish services from orchestration19:43 ↗
- Control behavior without adding unnecessary turns23:43 ↗
- Network distance, tool calls, and large context29:06 ↗
- Prepare credentials and understand the adapter boundary34:36 ↗
- Build the minimal Python agent40:40 ↗
- Replace the room transport and connect real clients52:32 ↗
- Test through voice, then try interrupting57:15 ↗
- Silence does not always mean the turn is complete1:00:34 ↗
- Synchronize visible words and assess offline speech1:03:23 ↗
- Word Wrangler: control what each agent can hear1:06:45 ↗