Subtitles for a finished stream, a highlight or a clip. Twitch links are fetched when Twitch allows it, which is less than half the time, so the dependable route is to download your VOD and upload the file.
How it works
In the Creator Dashboard open Content, then Video Producer, and choose Download on the stream. Do it soon: Twitch deletes past broadcasts after a set number of days that depends on your account.
A multi-hour VOD is several gigabytes. Files up to 5 GB go through the upload page; pulling the audio out first makes the upload a few minutes instead of an hour.
Use the SRT in your editor when cutting highlights, or upload it with the video to YouTube.
Link or file
Over the last two months fewer than half of the Twitch links pasted here could be fetched. The usual reasons are on Twitch's side: subscriber-only VODs, VODs that expired between copying the link and pasting it, and Twitch refusing automated downloads. A link that works is the least effort, so try it; when it fails the message says so within seconds and the file route below always works for your own streams.
One more difference: because stream recordings run for many hours, a pasted VOD link asks for a pack before the fetch starts. An uploaded file gets the free preview like any other recording, and so do clips.
For a long stream, extract the audio before uploading. The free extract audio from video tool does it in the browser, and a six-hour stream becomes an audio file of a few hundred megabytes. Cue times are unchanged, so the SRT still fits the full video.
A stream that is still live cannot be transcribed; wait for it to end. Kick replays behave the same way: see Kick VOD to SRT.
What stream audio does to a transcript
Streams are harder than a podcast, and the subtitles show it in predictable places:
The chapters that come with the transcript are a quick way to find the moments worth clipping in a long VOD.
Sample output
1 00:00:34,680 --> 00:00:46,500 fourscore and seven years ago our fathers brought forth upon this continent a new nation conceived in in liberty, and dedicated to the proposition 2 00:00:46,500 --> 00:00:50,300 that all men are created equal. 3 00:00:50,300 --> 00:00:57,500 Now we are engaged in a great civil war, testing whether that nation, or any nation so conceived 4 00:00:57,500 --> 00:01:01,620 and so dedicated, can long endure. 5 00:01:01,620 --> 00:01:07,920 We are met on a great battlefield of that war we have come to dedicate a portion of that field 6 00:01:07,920 --> 00:01:14,240 as a final resting place for those who here gave their lives that their nation might live 7 00:01:15,360 --> 00:01:18,800 it is altogether fitting and proper that we should do this
Real output, unedited: a public-domain LibriVox reading of the Gettysburg Address (2 min 41 s) transcribed here in October 2026. The repeated word and the missing capitals are what came back; you would fix those on the transcript before downloading.
Before you rely on it
Each cue is a phrase from the transcript with its start and end time. Most run a few seconds. A passage spoken without a pause can come out as one long cue of up to about half a minute, so split those in your editor if your platform wants two short lines at a time.
Cue times are taken from the recording itself, so the file lines up with the video or audio you uploaded. If you trim the start of the video afterwards, shift the cues by the same amount in your editor.
On public test recordings, about 7 words in 100 differed from the reference for clean English read speech, between 4 and 9 in 100 for Spanish, Portuguese, Russian, French and German, and about 21 in 100 for Hindi. Noisy rooms, music and crosstalk do worse. Method and every number: the benchmark.
Open the transcript, correct a misheard name or number on its line, and the SRT and VTT you download carry the correction with the timing unchanged.
You get the subtitle file. Fonts, colours, position and captions drawn into the picture are done in your video editor or player, which reads this file.
Up to 10 hours or 5 GB per recording. The box on this page takes files up to 500 MB; larger ones continue on the upload page. 99+ languages, detected automatically.
Pricing
The preview is free. When the full transcript needs a pack, one-time packs are $4 for 500 minutes, $8 for 1,000 minutes, $12 for 2,000 minutes, $24 for 5,000 minutes. Credits never expire, which comes to about $0.29 to $0.48 per hour of audio.
| Recording | $4 or $8 pack | $12 pack | $24 pack |
|---|---|---|---|
| 30-second clip | $0.01 | $0.01 | $0.00 |
| 2-hour stream | $0.96 | $0.72 | $0.58 |
| 6-hour stream | $2.88 | $2.16 | $1.73 |
Figures are the share of a pack each recording uses. A button opens secure checkout for that pack. Every export format is included; full details on the pricing page.
FAQ
You can try. Twitch links are fetched when Twitch allows it, which in the last two months was fewer than half of attempts, and a VOD link asks for a pack before the fetch starts. If the fetch fails, download the VOD from the Video Producer and upload the file.
Up to 10 hours or 5 GB per file. Streams of two to six hours are routine; extracting the audio first makes the upload much faster.
No. It transcribes finished recordings. Once the stream ends, use the VOD.
Loud game audio or music under the voice makes speech harder to recognise. Recording the microphone on a separate audio track and uploading that track gives much better subtitles.
Yes. Upload the VOD or highlight to YouTube, then in YouTube Studio open Subtitles, add a language and upload the SRT with timing.
Every page in this set uses the same upload box and the same export. All subtitle pages
Free preview on uploaded files; VOD links need a pack first. Up to 10 hours per file; packs from $4.