Add Subtitles to Songs Like a Pro | Descript Tutorial for Creators | Humanitarians AI
A hands-on Descript walkthrough for adding synced, styled captions to a song, including how to fix transcription errors without breaking the timeline.
Automatic transcription gets you most of the way to a captioned lyric video, but "most of the way" still leaves errors, timing drift, and words that never land where they should. This tutorial from Deepali Kashab, part of the Lyrical Literacy project's production process, walks through exactly how to take a song from raw upload to a properly captioned, exported video in Descript, including the specific fixes needed when the automatic transcript gets something wrong.
Starting with upload and automatic transcription
Descript is available both as a web app and a desktop application, and the workflow starts the same way in either: create a new video project, choose to upload a file, and select the song or music video from your desktop. Once uploaded, Descript automatically transcribes the audio, listening through the track and generating captions on its own. This step can take a little while depending on the length of the file, and once it finishes, the full transcribed lyric text appears on the left side of the screen, ready for review.
Checking the transcript before trusting it
The first real task after transcription finishes is not adding captions, it is proofreading them. Automatic transcription makes mistakes, and the only reliable way to catch them is going through each line against the actual audio. In the walkthrough, the transcript initially rendered a lyric as "under a tree" when the actual line was "under a banyan tree," a small but real error that would have shipped into the final captions if left unchecked.
Adding and styling captions
With the transcript reviewed, adding the visible captions is a short process: on the right side of the interface, a captions option lets you choose a font style for how the text displays over the video, with the tutorial settling on a typewriter style. Once applied, playing the video back confirms the captions are appearing and syncing roughly with the audio, which is the checkpoint before moving into detailed correction.
Correcting words without breaking the timeline
This is the part of the workflow that matters most for anyone new to Descript: how you fix a wrong word. Double-clicking a word brings up a correction option, where you can type in the right word and confirm it. Critically, you should never try to type a brand new word directly into an empty space, since that does not work reliably. Instead, when you need to insert an additional word, like adding "banyan" before "tree," you select the existing adjacent word, use the correct option, and add the missing word as part of correcting that existing word. Deleting a word outright is also risky: removing a transcribed lyric leaves a visible dashed line in the timeline and creates a jump or small delay when the audio plays back, since the tool assumes you deleted something that was there for a timing reason. The safer approach is always to add or delete relative to an existing word rather than manipulating empty space directly, and to avoid inserting more than three or four words in a single spot at once, since larger insertions are more likely to shift the timeline out of sync.
Fixing timing drift
Even with correct words, timing can still drift out of sync with the audio, a caption might display the right text but appear too early, too late, or not stretch to cover a longer sung word like "fortune." The fix here does not touch the underlying video or audio at all, only the caption timeline: you place your cursor on the caption, hold Command on Mac or Control on Windows, and drag it to the exact timestamp where it should appear, adjusting both when a caption starts and how long it lasts until playback confirms it lines up cleanly with the sung line.
The three-part workflow, summarized
The entire process reduces to three checks repeated line by line: confirm the transcribed lyric text is accurate, apply and style the captions with a chosen font, and sync the caption timeline to the audio so there are no delays or jumps. It is more manual than a fully automated captioning pipeline, but it is what produces captions that actually sync to a song rather than approximately following it.
Exporting the finished video
Once every line is corrected and synced, exporting happens from the export option in the top right corner. Descript offers two paths: a local export that saves the finished video directly to your computer, or a Descript web link that can be shared with anyone to view the same video inside Descript itself. For the local export, the recommended settings are the current composition, maximum resolution, and high quality, with no other adjustments needed before exporting.
Key takeaways
- Descript automatically transcribes uploaded audio into captions, but the transcript needs a manual review pass against the real audio before you trust it.
- Correct a wrong word by double-clicking it and using the correct option, rather than typing directly into empty space.
- To insert a missing word, correct an adjacent existing word rather than adding a brand new one, and avoid inserting more than three or four words at once.
- Deleting a transcribed word outright can create a timeline gap and cause a playback jump; add or delete relative to existing words instead.
- Fix timing drift by holding Command (Mac) or Control (Windows) and dragging a caption to the correct timestamp on the timeline, without touching the underlying video or audio.
- Export locally for a file on your computer, set to current composition, maximum resolution, and high quality, or generate a shareable Descript web link instead.
Try it yourself
This is a practical, repeatable workflow for anyone adding captions to a lyric video, educational song, or sing-along, whether for accessibility, classroom use, or general engagement. It is part of the production process behind Humanitarians AI's Lyrical Literacy program, which uses AI-generated music to create personalized educational content.
Full transcript(auto-generated, with timestamps)
[0:06]Hi everyone, I am Deepali Kashab and uh today I'll be uh talking about this tool called Descript. Uh this is available on a website and also an application. So you're free to use whichever suits you well. Uh we use this tool for the lyrical literacy project. uh right now with for limited features only but um which I'll be explaining in this video but we will add on to it later. Uh so let's dive deep into the tool right now. Uh what you're seeing on the screen is uh how the tool looks like the homepage. Uh the first step that we need to do is when we working with a video we need to
[0:45]Uh upload the video. Uh so we go to the new project option here on the top right corner. We say pro video project. So here you have an option to upload a file. So we'll say upload a file. Go to desktop. Okay. I'll just take one of the video that I have here. So what it does is it uploads the video along with it. It transcribes the video. So that is it goes through the audio of the video and then tries to read them and give us the captions by itself. This could take a while. Let's wait until it's uploaded. So now we have completed the
[1:51]Transcription. So you can see here once the transcription is completed all the lyrics of the video uh comes on the left hand side of your screen. Um what we have to do next is to go through the um lyrics that it has transcribed and check if it actually is right because sometimes there can be some errors which we need to manually correct them and that's what we'll be doing next. Okay. So we are almost done. Um just a couple of seconds more. Okay, perfect. So now it's uploaded the uh video for us and it has also transcribed the lyrics from the video. Uh the first thing that we need to you
[2:49]Usually do is go through each line of the lyric here on the left hand side of the screen and make sure that it's all right according to the audio in the video that we have uploaded. And the second thing that we do usually with this tool is to add the captions for the video. So let me first tell how to add the captions. Uh on the right hand side you have a couple of options here and if you just look down there is an option for captions here. Just click on that and you have a couple of options for what are the fonts of the captions. So
[3:28]We usually go with the typewriter. So let me choose that. It's applying. Okay. So once the um captions are already there. So now let's just try to play the video and see if the captions are actually on the video. Uh I'll just alter it so that it's more clear. Okay, let me play. >> Okay, perfect. So we can see that the um lyrics are captioned and it's appearing on the video. So the next thing as I already said we need to check if the uh lyrics are right if each word is uh correctly transcribed. Uh so there's no other way we need to do this manually.
[4:15]So let's just listen to the audio again. >> Okay. So uh we understood. So it said in a temple tucked under a banyan tree. But you can see uh that it's not um right here. So let's just edit it. So what you need to do when you need to correct a word is just doubleclick on it. Sorry. Then you get a couple of options here. So we say correct here. You can just give the right word that you need to insert and say correct. We also notice that it says banyan tree and not just tree. So uh if I try to just add a word here that doesn't work.
[5:05]So what I usually do is click on this word say correct and just add banyan tree and say correct. So that inserts the word there. So usually when we need to add a word behind an already existing word or add a word after we cannot type it directly. We just choose the word that's there and then correct it there. So in case you try to like insert a new word it usually let me just show you. So imagine I want to add or I just want to delete this. So I just deleted that word there. So now you can probably see that there is a vertical
[6:02]Dashed line here. So what happens is it just removed one of the lyric that it transcribed. So now when we play the audio you can see there will be a jump or a small delay. this is not something that we want. So we usually don't try to alter the lyric that it has transcribed but instead we try to add a word or um uh delete something uh using the previous or the word after it. >> So did you uh hear that there was a jump? So we don't want that. Let me just Yeah. So uh this is pretty much how we uh edit the whole song. We need to go line by
[6:50]Line, make sure each line is right and then uh make the fixes wherever it's required. Um there is one other um problems usually that happens that is um let me just try to give you an example. Let's play from here. Three. Make it two. Make it three. >> Okay, let me just correct this one line and then we'll tell you what's the problem here. So, it said Make it to make it free. And with time forune.
[8:10]Okay, so this last line of the song is now fixed. Um, another hint is or like when you're trying to add new words, um, don't try to add more than three to four words um, in the same place. uh you can pick the next word and then try to correct it there just so the timeline doesn't get shifted or there's no issue in the song. So another problem that we usually encounter is something that I'll show now. Let me just play it. with time forunate. >> So if you noticed I added good fortune there will be but the caption never appeared there. That's probably because
[9:10]Um in the timeline there is a shift in the place where the captions added. So we need to fix that. So, we don't usually mess with the timeline. Let's see. Yeah. So, here this is the video. This is where the captions are added. And this is the audio. So, we did not touch the video or the audio. So, we just work with the timeline of the captions and sync it to the right minute or the right time that we need it to appear. So, where is it? Good fortune. And fortune's pretty long. Yeah. So there'll be here. So we need to edit this. So you just put your cursor
[10:11]Um to where you see it. And uh I think in u Mac the option is command and in um Windows I think it's control. So you just hold it and then you pull it to the exact time stamp where you need fortune to be appearing. Okay. And similarly we got a pull d as well.
[11:23]We are almost done. Okay, let's now just try to play it and make sure it's right or if we need more changes there. We end with time forune. Okay, you see there's still some changes required. We need to pull up the duration of fortune. So, let's do that. Okay, let's try to play it again and see if it works this time.
[12:38]Okay, perfect. So, we almost got it, but you get the whole point. How to make changes and sync the captions to the audio um in the video. So, there are three things that we need to do here. One is to make sure that each word in the lyric that has been transcribed is right. uh we need to add the captions. Choose the correct font and add the captions. And the third thing is make sure that the timeline of the caption syncs with the audio and there are no delays or jumps that occurs throughout the video. So this is one reason why we use a script currently,
[13:19]But I'm sure maybe in the future we'll add for more features or use up the tool more extensively. Lastly, uh I'll just show how we'll export the video that's fully completed after adding all the captions. Uh all that we need to do is you see the export option on the top right corner. Go there. Um we can we have two options here. We can either do a local export um that is it'll uh put the video in your uh desktop or whichever location you choose to your computer. uh or we can even do a descript web link. Uh so if we do a descript web link, we can uh share
[14:01]This link with anybody and they can access the same video on the descript. Uh we also usually make sure that the composition is set to current composition and resolution set to max and the quality is at high. Uh and we don't do any other changes here. And then we just say export and it puts it to your computer. Thank you.
More videos
2:08Bridging the Pixel Gap in Browser Automation.
2:23How One Narrow Safety Rule Can Make an AI Less Safe Everywhere Else.
2:04Why splitting a chunk from its document makes it retrieve for the wrong question
4:20Three You Can Take Back. One You Can't.
2:21Why a 50-turn agent pays for the same screenshot 35 times unless it caches the pixels
1:53