Transcription Reviewer - Clinical Conversation Data

About this role

## Overview We are building a set of verbatim human transcripts from real, de-identified clinical phone calls, used as reference data for evaluating and improving speech recognition systems. Transcribers type each call from scratch, by ear, with no machine draft. You would review their work. Accuracy matters far more than speed on this project, and that applies to reviewers most of all. ## What you will do - Listen to the call, read the transcript against it, and decide whether it meets the standard. - Mark each transcript approved, or send it back with specific, timestamped notes. Feedback goes inline in the transcript document, plus a short summary. - You are checking five things: - **Coverage.** Does the transcript run to the end of the call? - **Accuracy.** Does it match what was actually said? - **Verbatim discipline.** Has the transcriber tidied up speech that should have stayed messy? Filler words, false starts, stutters, repetitions and short acknowledgements all belong in the transcript. - **Flagging.** Are uncertain clinical terms flagged rather than guessed? - **Formatting.** Turn-level timestamps, speaker labels, and a fixed vocabulary of bracket tags for things like laughter, crosstalk and inaudible passages. - Help set the standard early. The first wave of reviews is where the conventions get established for everyone on the project. ## What we are looking for - Prior experience reviewing, editing, proofreading or QA-ing other people's transcripts, with a track record of being rated well for it. - Background in medical transcription, clinical documentation, medical scribing, closed captioning, or legal and court transcription. - AHDI credentials such as CHDS, RHDS or CMT are a strong plus, not a requirement. - The kind of attention that catches what others read straight past: a plural that should be singular, a missing hyphen in a drug name, a timestamp that has drifted a few seconds. - Comfort with detailed written specifications and consistent formatting conventions. Consistency between reviewers is the single biggest quality factor on this work. - Native or near-native English listening comprehension, including across a range of accents. - A quiet working environment and over-ear headphones. Laptop speakers are not sufficient for this audio. ## How the work runs - Remote and hourly. This is a pilot, with a larger phase behind it if it goes well. - Everything is done by ear. You may not run the audio through any transcription tool or AI service, at any stage. This is the one rule that ends the engagement. - All audio is de-identified before it reaches you and is handled under a strict data agreement. It stays inside the project's tools. ## Note on content This audio consists of real clinical conversations. Content may reference medical conditions, symptoms and treatment, including topics some people may find distressing. Please apply only if you are comfortable working with material of this kind.

Eligible applicant countries

This role accepts applicants from:

• USA

Skills & domains

• ai-training
• rlhf
• sme
• annotation