How do you transcribe a long audio file, such as one over 30 minutes?

To transcribe a long audio file, cut it into parts at sentence breaks, transcribe each part, then join the text and check the joins against the recording. Many tools accept a file only up to a set size or length, so the split is often required. The machine part is quick, and the proofreading takes most of the time.

Long recordings are processed in pieces, so split them at sentence breaks

Whisper's models are trained on 30-second chunks of audio and cannot take a longer input as a single piece ¹. That is why long recordings are handled in parts. If you split the file yourself, cut between sentences. OpenAI advises against splitting a recording in the middle of a sentence, because the cut can remove context and reduce accuracy ².

After each cut, check the first few words of the new part against the audio. A cut through a word or a name is an error that is easy to miss when you read the joined text.

Check the upload limit before you split

Each service sets its own limits, so read them before you start. OpenAI's Transcriptions API accepts files up to 25 MB, and larger recordings must be compressed or split into chunks ³. Compressing to a format such as mp3 or m4a usually makes a long file much smaller.

Limits also depend on the route you use:

If your recording is longer than a single request allows, use a batch or file route rather than a short-audio route. A batch job can wait in a queue, so start it early.

Speed: a fast model processes an hour of audio in minutes

A long recording takes longer to process, but a fast model is quicker than you might expect. The turbo Whisper model runs at about 8 times the relative speed of the large model ⁶. In one 2025 test of a 7-minute-31-second podcast episode, Whisper Large V3 Turbo took 40 seconds. Scaled to an hour of audio, that works out to about 5.3 minutes of processing ⁷. That is one test on one episode, so treat it as an estimate.

Manual transcription takes far longer. A Claremont Colleges oral history guide says that by some estimates one hour of audio or video can take 4 to 9 hours to transcribe, depending on the subject, the number of speakers and the audio quality ⁸. The gap between a machine draft and a person typing is the main reason to start with a tool.

Audiobooks and lectures: split by chapter or topic

A long audiobook or lecture has natural breaks. Split by chapter or by topic, and keep each file under the limit. This also makes the proofreading easier, because you can work one chapter at a time and track where you stopped. For lectures, the step-by-step method is in how to transcribe a lecture.

Names of people and places, and technical terms, are the first things to check in long material. Check them once for each part. If a term keeps coming out wrong, the custom vocabulary feature in some tools can help, though it is not available in every tool.

It depends: a formal record needs a full check

A personal recording needs one careful read. A recording that will be quoted, used in research or kept as a legal record needs a full check against the audio, and a human review if the stakes are high. When several people speak, you also need to check who said what. The guide to timestamps and speaker labels covers that step.

What to do

1. Measure the recording's length and file size, and check your tool's limits.
2. Compress the file if it is over the limit, then split it at sentence breaks if it is still too large.
3. Transcribe each part, and keep the parts in order.
4. Join the text, then check the first and last words of each part against the audio.
5. Proofread names, numbers and dates, then check any quiet passages for text that was never spoken.

The full method is in our guide to transcribing audio to text. For the basics, see what it means to transcribe audio, and for the formats most tools accept, see which audio formats can be transcribed.