How Long Does It Take to Transcribe a Video?
Measured on real jobs, not estimated: captions come back in milliseconds, AI transcription has a 10-second median, and the slow tail has three specific causes.
Short answer
If the video already has captions, the transcript comes back in well under a second. If it does not, the audio has to be downloaded and transcribed: the median job finishes in about ten seconds and nine in ten finish within thirty. Multi-hour recordings are the honest exception and take minutes.
Short answer, from our own production traffic: if the video already has captions you will have the text in well under a second. If it does not, the median is about ten seconds and nine jobs in ten finish inside thirty.
The longer answer is more useful, because the cases that take minutes take minutes for reasons you can predict before you paste the link.
There are two completely different paths
Most tools blur these together. They should not be.
The caption path. Some platforms publish a caption track, and where one exists the job is a fetch and a parse. There is no audio download and no speech recognition. Measured over recent production traffic, the median request on this path is 0.01 seconds and the 90th percentile is 1.24 seconds. The variation is almost entirely the platform's own response time, not ours.
The AI path. When no caption track exists, the audio has to be downloaded and run through speech recognition. This is the path that takes real time, and it is where every number below comes from. Measured over 72 completed jobs: median 9.9 seconds, 90th percentile 27.9 seconds, 99th percentile 203.8 seconds.
Some platforms never have captions and always take the second path — Facebook, Instagram, Google Drive, Dropbox, Kick and uploaded files among them.
Where the time actually goes
Breaking the AI path apart, over 552 successful audio extractions in a week: median 6.2 seconds, 90th percentile 99.1 seconds, 99th percentile 378.6 seconds, slowest 492.5 seconds.
That distribution is the whole story. The median job spends about six seconds fetching audio and the rest transcribing. The slow tail is almost entirely download, not recognition, and three things put a job there.
1. The video is long. Obvious, and the least interesting. Our own mix: 2,058 finished jobs under five minutes, 2,119 between five and thirty, 1,026 between thirty and sixty, 1,234 between one and three hours, and 100 over three hours, the longest averaging five and a half hours.
2. The platform publishes no audio-only version. This is the one people never predict. On most platforms a downloader can ask for audio and get a small file. On some — Wistia, Rumble, Dailymotion, naver TV among them — every available version is video and audio welded together, so getting the audio means downloading a video file. If the tool asks for the best available version it downloads the largest file on the site to reach audio that is bit-for-bit identical in the smallest one. We measured that penalty at 123 times on Wistia, 22 on naver TV, 13 on Dailymotion and 10 on Rumble before fixing it by always requesting the smallest version that still carries sound.
The reason it matters to you: when a tool is inexplicably slow on a short video, this is usually why, and it is invisible from the outside.
3. The platform is stalling rather than refusing. A rate limit that returns an error is fast. A rate limit that simply stops sending bytes looks identical to a slow connection until the timeout fires. That is the shape behind most jobs in the 99th percentile.
What this means in practice
- A short video with captions is effectively instant. If a tool takes thirty seconds on a ten-minute YouTube video, it is transcribing audio it did not need to download.
- A one-hour video with no captions is a one-to-two minute job, not a ten-minute one, unless it hits the audio-only problem above.
- Multi-hour recordings are the honest exception. A five-hour livestream VOD is a genuinely large download before any transcription starts. Transcribing a full livestream VOD covers what to expect, and long videos and multi-hour podcasts covers the same ground for recorded content.
If a job is taking much longer than this
It is usually one of three things, in descending order of likelihood.
1. The video is longer than you think. Livestream VODs are the usual culprit; a stream you watched for twenty minutes may be a six-hour recording.
2. The platform is rate-limiting, and the tool has not admitted it yet. The tell is that a second attempt a few minutes later behaves completely differently. Why a TikTok transcript fails and then works is the clearest example.
3. The job already failed and is waiting on a timeout. A well-built tool should tell you rather than leave a spinner running, and should return your credits when it does.
A note on what we measured
These are wall-clock numbers from production on 15 September 2026, not a benchmark run on a fast machine with a favourable video. They include every real job: the phone on a bad connection, the platform having a slow afternoon, the six-hour stream.
The caveat worth stating: our servers are in one place, and the distance between them and the platform is part of every number above. Your own upload speed does not matter for a URL, since we fetch the video, not you — but it is the entire story if you upload a file instead.
Frequently asked
Why is transcribing a short video sometimes slower than a long one?
Usually because the platform publishes no audio-only version. Every available file is picture and sound welded together, so reaching the audio means downloading a video. We measured that penalty at 123 times on Wistia, 22 on naver TV, 13 on Dailymotion and 10 on Rumble before fixing it by always requesting the smallest file that still carries sound.
How long does a multi-hour livestream VOD take?
Minutes rather than seconds, and almost all of it is download rather than transcription. Across a week of successful extractions the median took 6.2 seconds, the 90th percentile 99.1 seconds and the slowest 492.5 seconds. Speech recognition itself is a rounding error next to fetching the file.