Question 1

How long does it take to transcribe 1 hour of audio?

Accepted Answer

Typing it yourself takes four to six hours per hour of audio, because you play, pause, type and rewind constantly. A human transcription service returns it the same day or the next. Speech recognition on your own machine takes a few minutes: whisper.cpp runs roughly four to twenty times faster than realtime on an ordinary CPU, and faster still on a GPU.

Question 2

Why is typing a transcript so much slower than the recording?

Accepted Answer

Because you do not type at the speed people talk. A meeting runs at 130 to 160 words a minute and sustained typing is closer to 40 to 60, so every minute of audio needs several passes: type, rewind, catch the word you missed, fix the name. Crosstalk and unfamiliar accents make it worse.

Question 3

Is local transcription as accurate as a paid service?

Accepted Answer

For clear meeting audio, a mid-sized Whisper model running locally is close enough that most people stop noticing the difference. A human transcriptionist is still better on heavy crosstalk, strong accents and specialist vocabulary. The gap costs you a dollar or two per minute and a trip through someone else's servers.

Method	Time	Cost
Typing it yourself Play, pause, type, rewind. Four to six minutes of work per minute of audio, worse with crosstalk or an unfamiliar accent.	4 h – 6 h of your work	free
A human transcription service You send the file and wait. Standard turnaround is same-day to next-day; rush costs more.	12 h – 24 h of waiting	$61 – $119
A cloud AI service Minutes, most of it spent uploading. Your recording sits on someone else's servers while it runs.	3 min – 12 min of waiting	$10 – $15
On your own machine whisper.cpp on the CPU you already have, roughly four to twenty times faster than realtime depending on the model. A GPU beats the whole range. Nothing is uploaded.	3 min – 15 min of your work	free

How long does transcription take?

Local recording tips, once in a while

Related

Stop doing this by hand