What is transcription?

Transcription is writing down exactly what is said in an audio or video recording. In AI training it produces the data that teaches speech recognition models to turn speech into text, and that helps voice assistants understand accents, background noise and specialist vocabulary.

What the work looks like

You listen to short clips and type what you hear, following a style guide: whether to keep filler words such as "um", how to mark overlapping speakers, how to tag noises and unclear words, and how to write numbers. Many projects also ask for speaker labels and timestamps. Language and dialect projects need native speakers, which is why transcription is one of the most international kinds of AI training work.

How it is paid

Transcription is usually paid per audio minute or per task. Medical and legal transcription, and rare languages, pay more because fewer people can do them accurately.

Example

A 30-second clip of a customer call includes a dog barking and two people talking over each other. The transcript labels Speaker 1 and Speaker 2, marks the overlap, tags [dog barking], and writes an inaudible word as [unclear] rather than guessing. See open Languages roles.