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

For an AI engine, one hour of audio takes seconds to a few minutes. For a person typing it by hand, it takes about four to nine hours. Those are the two ends of the range, and the gap is the main reason people ask. The exact time depends on the engine, the audio and whether you count only the first draft or a checked, finished transcript.

An AI engine returns the text in seconds to minutes

Speed is measured as an inverse real-time factor, the amount of audio processed for each second of compute time. On the Open ASR Leaderboard's English short-form track, Whisper Large v3 reached 146 ¹. That works out to about 25 seconds of compute for one hour of audio ². The leaderboard measures throughput on batched audio, so a single file on your own computer will take longer, but the order of magnitude holds.

A single real test shows the same spread. In a 2025 9to5Mac test of one 7-minute-31-second episode, Apple's speech API took 9 seconds ³. Scaled to one hour, that is about 72 seconds ⁴. Whisper Large V3 Turbo took 40 seconds ⁵ on the same episode, which scales to about 5.3 minutes for an hour ⁶. The engine you choose changes the wait by a factor of four or more, though both were well under real time.

Typing by hand takes four to nine hours

Manual transcription is slow because a person has to listen, pause, type and replay. A Cornell interview project guide tells interviewers to expect to spend 4 to 6 hours transcribing per hour of recorded interview ⁷. A Claremont Colleges oral history guide says that by some estimates one hour of audio or video can take 4 to 9 hours, depending on the subject, the number of speakers and the audio quality ⁸.

Typing speed sets a floor. A one-hour recording at 150 words per minute holds about 9,000 words, which takes about 175 minutes to type at the 51.56 words per minute average from a large online typing study ⁹. Real transcription runs slower than pure typing, because the replay and the pauses take time. Those two sources are planning estimates rather than measured averages, so use them to plan rather than to promise a deadline.

A human service takes hours, and the price includes the wait

A professional service sells the finished file, not the first draft. Rev's human service promises 99 percent accuracy or better, in 12 hours or less, at $1.99 per minute ¹⁰. The 12-hour figure is the window the service states for delivery, so it is the right number to compare with an AI engine's seconds.

It depends on what counts as done

There are three different times, and people often mix them up:

• The raw AI draft, which arrives in seconds to minutes.
• A checked draft, where you listen to the risky parts and fix names, numbers and quiet passages. That can take a good fraction of the recording's length, depending on the audio.
• A finished, certified transcript, which needs a person to take responsibility for the text. Certified court records are a different case from a meeting note.

For a meeting, the checked draft is usually enough. For a record that must be exact, plan on the human time, not the engine time.

What to do

1. Decide the deadline and the standard. A draft for your own notes needs only the first time. A record for others needs the checked or certified time.
2. Run the engine first. For most recordings, an AI draft takes minutes, so you can see the audio quality before you commit to more work.
3. Budget review time. Checking takes longer on poor audio, with many speakers or with specialist terms, and much less on a clear single-speaker recording.
4. Use a service for the rest. If the deadline is tight and the audio is hard, a human service delivers in hours rather than days.

For accuracy figures, see how accurate AI transcription is, and for prices, see how much transcription costs per minute. The guide on human versus AI transcription compares the routes side by side. Our overview of transcription covers the rest of the subject. The measure of speed in the leaderboard is explained in the real-time factor entry, and the human service's delivery promise is a turnaround time.