Teaching a Computer to Speak | Voice AI in the Real World

Voice AI sounds simple until you try to deploy it at scale — across airports, accents, languages, and thousands of employees at once.

In this episode of ServiceNow Insights, host Bobby Brill sits down with Midam Kim, an ML engineer and linguist at ServiceNow, to unpack what it actually takes to build voice AI as an enterprise product. From the out-of-vocabulary problem (why AI still struggles with names) to why turn-taking in conversation is a linguistic skill most people never think about, Midam breaks down the human science behind the technology.

In this episode:

- Why voice AI is replacing typing as the default way to interact with enterprise systems

- The difference between building for employees (B2B) vs. building for their customers (B2B2C)

- Why an airport is one of the hardest possible environments for voice AI — and what ServiceNow does about it

- The "out-of-vocabulary" problem: why AI still struggles with names, accents, and rare expressions

- Why ServiceNow's secret sauce is hiring linguists, not just engineers

- The linguistic framework behind every voice interaction: sounds, words, and turn-taking

- Why voice AI is like teaching a kid to speak for the first time

Chapters

00:00 — Welcome to ServiceNow Insights