Ann: Successful Completion of A$285 Million Placement, page-93

[ I worked for Appen in various senior roles for 13 years, left last year. I invested a few years prior to the IPO. ]

All the big tech companies (car makers license them from others, almost all use Nuance) build their own recognition engines, but none of them collect and annotate all their own training data. Many used to do one language by themselves (usually US English) during their internal development, but now even that is too hard. Training a modern speech recognition engine requires somewhere between 1000 and 10,000 hours of annotated (transcribed, including marking things like breath noise and background noise) speech, in a variety of environments.

Appen works for all the big players to collect data in a huge number of languages around the world, and then annotate it, but Appen doesn't produce any of the technology for them.

Essentially Appen is a linguistic consultancy combined with (several, after the acquisitions) crowd-sourcing platforms.