Human transcripts: HappyScribe $2.00, Rev $1.99 a minute

Human-verified transcripts run about $2 a minute: HappyScribe lists proofreading from $2.00 per minute and Rev AI lists human transcription at $1.99 per minute. A machine transcript on Sume is $0.01 per audio minute plus inspect compute, so a human pass is about 200 times the transcript line. Whether that is worth it depends on the cost of an error, and Sume does not provide the human step.

The numbers from the vendor pages

HappyScribe says its proofreading starts at $2.00 per minute, with a 5 percent discount on Business at $1.90, and delivery in 4 hours or more. Rev AI lists Human Transcription at $1.99 per minute on its pay-as-you-go page.

A hybrid that costs little

The cheap pattern is to run the machine transcript everywhere and send a human only the clips where a mistake is expensive: legal, medical, brand names in paid ads. If one in ten hours goes to a human at $120 and the rest cost $0.60 each, a 10-hour month costs about $125.40 before inspect compute, against $1,200 for all-human.

Steps: build the hybrid

• Import each clip to media.sume.com with POST /v1/media-imports.
• Run video inspect with transcribe set to true and a duration_seconds hint.
• Flag clips that contain names, numbers or regulated claims.
• Send flagged clips to your human vendor and keep the rest as machine text.

What Sume does not do

Sume has no human transcription or proofreading service and does not publish an accuracy figure in the docs I read. The machine transcript can mishear names; treat it as a draft where errors are costly.

How to choose which clips get a human

A simple rule keeps the human spend small. Send a clip to a person if it will be published under your name with exact quotes, if it contains numbers a viewer could act on, or if the speaker has an accent or jargon your machine pass got wrong on a test clip. Run five clips through the machine step first and read them. If a clip needs more than two corrections per minute, pay for the human version; otherwise ship the machine text and fix the odd error yourself.

Turnaround and language notes

HappyScribe lists human delivery in 4 hours or more and 65-plus languages, while Rev's page gave a price but I did not extract a turnaround time. Sume's transcript accepts a language_code hint such as en or ko, or detects the language itself. If your deadline is minutes rather than hours, the machine pass is the only option here, and a human review is something you schedule afterward.

Bottom line

Pay about $2 a minute only for the audio where accuracy is a legal or brand risk. For everything else a cent-a-minute line, measured on a real clip first, leaves your budget for the clips that matter.