How to Transcribe Large Audio Files and Long Videos
By Eman Khoubian, Founder & CEO · Updated
Upload the whole file. WhisperAI takes recordings up to 5 GB and up to 10 hours on every plan, so a long interview, lecture or webinar goes up in one piece, with no splitting and no converting. Just make sure your plan has the minutes: a 3-hour recording uses 180.
Only two kinds of file need work first: one bigger than 5 GB, which is usually a video, and one longer than 10 hours. Both fixes keep your timestamps usable.
What is WhisperAI's maximum upload file size and length?
5 GB per file and 10 hours per recording, on Free, Premium and Business Pro alike. Go over 10 hours and we skip the file at upload and ask you to split it. Go over 5 GB and the upload is refused.
Audio can be MP3, M4A, WAV, FLAC, AAC, OGG, OPUS or WMA. Video can be MP4, MOV, MKV, WEBM, M4V, MTS or MXF, among others. Upload a video as it is: you don't need to pull the audio out first.
If a recording is longer than the minutes you have left, we refuse it. It isn't transcribed halfway. The free plan has 5 minutes a month. Premium has 120 minutes a month for $14.99, enough for one 2-hour recording. Business Pro has unlimited minutes for $24.99 a month, which is what a 6-hour deposition or a weekly 3-hour podcast needs.
A big upload goes straight from your browser to cloud storage, a piece at a time. If your connection stalls on a piece for a minute, that piece is sent again from the last byte that arrived, so bad hotel Wi-Fi slows a 4 GB upload down without restarting it. Leaving the page does cancel it, and the browser warns you first. Once the upload finishes, the transcription runs on our servers, and you can close the tab or shut the laptop.
Can you transcribe large audio files over 2 GB?
Yes, up to 5 GB. And for most audio, the 10-hour limit hits long before the 5 GB one. It comes down to the bitrate, which is how much data each second of sound takes:
- MP3 at 128 kbps is about 57.6 MB an hour. A 10-hour MP3 is about 576 MB, and 5 GB would hold about 93 hours. Length is your limit.
- WAV at CD quality (44.1 kHz, 16-bit, stereo) is 1,411,200 bits a second, about 635 MB an hour. 5 GB fills at about 8 hours 27 minutes, so a 2 GB WAV is a little over 3 hours.
- Video counts picture and sound together. Every 1 Mbps of bitrate is 450 MB an hour. At 2.5 Mbps, 5 GB holds about 4 hours 46 minutes. At 8 Mbps, about 1 hour 29 minutes.
How many hours fit in one upload
To check your own file, divide its size by its length. 5 GB over 10 hours is about 537 MB an hour. If your file uses more than that, it hits 5 GB before 10 hours. If it uses less, the 10-hour limit comes first.
How to transcribe a very large video file over 5 GB
If your video is over 5 GB, most of that is picture, and the transcript only needs the sound. Copy the audio track out with ffmpeg, a free command-line tool, and upload that:
Copy the audio track out of a video, without re-encoding it
ffmpeg -i webinar.mp4 -vn -c:a copy webinar.m4aFrom the ffmpeg documentation, ffmpeg 9.0.
The audio keeps the video's clock, so every timestamp in the transcript points at the right moment in the video.
If ffmpeg says it can't write that codec into an M4A file, turn the sound into an MP3 instead:
Fallback: re-encode the audio as a 128 kbps MP3
ffmpeg -i webinar.mp4 -vn -c:a libmp3lame -b:a 128k webinar.mp3From the ffmpeg documentation, ffmpeg 9.0.
Even a 9-hour video becomes an MP3 of about 520 MB.
How to split audio longer than 10 hours
A full-day conference or a 24-hour livestream goes up in parts. Keep every part under 10 hours, and cut in a pause so no sentence is split across two files.
Split a 14-hour recording into two 7-hour parts with ffmpeg
ffmpeg -i conference.mp3 -t 07:00:00 -c copy part1.mp3
ffmpeg -ss 07:00:00 -i conference.mp3 -c copy part2.mp3From the ffmpeg documentation, ffmpeg 9.0.
To cut somewhere else, change 07:00:00 in both lines. Neither command re-encodes the audio. Before you upload, check each part's length. This prints it in seconds. 36000 seconds is 10 hours, so you want a smaller number:
Check a part's length
ffprobe -v error -show_entries format=duration -of csv=p=0 part1.mp3From the ffprobe documentation, ffmpeg 9.0.
Rather click than type? Audacity does the same job. Open the recording, drag across the first part, open Export Audio from the File menu and set the export range to Current Selection. Then select the rest and export again. Save the parts as MP3, not WAV, or a long part can still come out over 5 GB.
Each part gets its own transcript, and each one starts at 00:00. To find a line in the full recording, add the time the part starts: 00:12:40 in part 2 of the example above is 07:12:40 in the original. The timestamps guide shows the same addition for lining a transcript up with a video timeline.
Speaker labels start over in each part too. Speaker 1 in part 2 may be a different person from Speaker 1 in part 1, so check who's who at the start of each part before you merge the text.
How to transcribe long audio saved as several files
Some recorders and apps save one long session as several files. If they add up to less than 10 hours, join them before you upload. You get one transcript with one clock, and one set of speaker labels from start to finish.
Put the file names in a text file called list.txt, one line each, in order:
Join the files without re-encoding them
ffmpeg -f concat -i list.txt -c copy session.mp3From the ffmpeg concat documentation, ffmpeg 9.0.
Each file picks up where the one before it ends. This works only when every file came from the same recorder with the same settings, because ffmpeg can't join files with different codecs this way.
Can OpenAI Whisper transcribe long audio files?
On your own computer, yes. The open-source model reads the whole file and works through it in 30-second windows. The cost is the setup: Python, ffmpeg, and for the large model a graphics card with about 10 GB of video memory.
OpenAI's API skips the setup, but it accepts files up to 25 MB. That's about 26 minutes of 128 kbps MP3, and anything bigger has to be compressed or split. WhisperAI is the no-setup route without the chopping: open it in your browser and upload the recording as it is.
Is there a transcription service with no file size limit?
Not among our closest rivals. Each one caps a single file:
- TurboScribe allows 30 minutes per file on its free plan and 10 hours or 5 GB on Unlimited.
- Otter imports files up to 5 GB, and its free Basic plan gives you just the first 30 minutes of each transcript.
- Rev's AI transcription takes files smaller than 5 GB.
So skip the hunt for a service with no limit and pick one your recording fits under. None of them takes a bigger file than WhisperAI does.
Before you send a whole day of audio, run 5 minutes of it through the free plan and check the speaker labels and the export you need. Then upload the full file.