Kdenlive and Shotcut can both turn speech into subtitles for free, using speech models that run on your own computer. In Kdenlive, set up a speech engine once under Settings > Configure Kdenlive > Speech to Text, then click Speech recognition on the subtitle track and select Process. In Shotcut, open View > Subtitles and click Detect speech… to run Speech to Text. Both put the result on a subtitle track you can correct, style, burn into the video or export as an SRT.
Both editors are free and open source and run on Windows, Mac and Linux. The steps below follow each project's own documentation as read on 19 October 2026: the Kdenlive 26.08 manual, and Shotcut's docs and release notes up to version 26.9.27. The last section is for when you want animated, styled captions for a vertical video rather than plain subtitles.
Kdenlive: speech-to-text setup and models
Kdenlive's speech features run on Python, so there's a one-time setup. From the Kdenlive manual's plugin chapter:
- Install Python 3. On Windows, download it from python.org and tick Add python.exe to PATH; most Linux systems already have it.
- In Kdenlive, open Settings > Configure Kdenlive and go to the Speech to Text section.
- Choose an engine. The manual says Whisper is slower than VOSK on a CPU but more accurate, and that it adds punctuation.
- For Whisper, let Kdenlive install the missing dependencies (about 2 GB the first time), then download a model. The manual recommends the turbo model, a 1.4 GB download.
- For VOSK, download a language model from the alphacephei page linked in the dialog and add it, or drag it into the models window.
- Optional: switch Whisper from CPU to GPU. The manual says this needs a GPU that supports CUDA.
Two notes from the manual. Models are large and can take a long time to download and unpack, during which Kdenlive may look frozen. And the Flatpak version of Kdenlive can have trouble installing speech models because of its sandbox; the manual gives a workaround.
Kdenlive: generate, edit, style and burn in subtitles
The speech-to-text chapter gives these steps:
- Add a subtitle track with the Edit Subtitle Tool icon in the timeline toolbar, if there isn't one yet.
- Click Speech recognition to open the Automatic Subtitling dialog.
- Choose what to analyze: the timeline zone (the default) or only the selected clips. Set the zone with the I and O keys.
- Select the Model. With Whisper, also set Language (Autodetect by default; pick the language if detection goes wrong) and Maximum character per line.
- Click Process. The subtitles are inserted on the subtitle track.
Whisper can also translate the speech into English as it transcribes, and Kdenlive can add SeamlessM4T for other target languages, a download of around 10 GB.
Edit. Type straight into a subtitle on the track or use the Subtitle Window. Drag a subtitle's ends to retime it, and use the scissors button to split one that's too long. Right-click the text box and turn on Auto Spell Check to catch typos.
Style. Kdenlive stores subtitles as ASS files, so styles carry font, size, colors, outline and position. Create and assign them in the Subtitle Manager (Sequence > Subtitles > Manage Subtitles > Styles), and use ASS override tags if you want one word in a different color.
Burn in. Open File > Render (Ctrl+Enter) and render as usual: subtitles are burned into the video by default. The render dialog's option Embed subtitles instead of burning them in keeps them as a separate stream instead, which needs a container such as MKV.
Export a file. Sequence > Subtitles > Export Subtitle File saves SRT or ASS for YouTube or a player.
Shotcut: generate subtitles from speech
Shotcut added its Subtitles panel in version 24.08 and Speech to Text in 24.10.29, built on whisper.cpp, a C++ port of OpenAI's Whisper. Since 25.07 the installer no longer includes a speech model; the Speech to Text dialog has a model downloader. Since 26.4 the dialog has a Use GPU checkbox, with Vulkan on Windows and Linux added to the existing Metal support on Apple Silicon.
From Shotcut's Speech to Text documentation:
- Put your video on the timeline.
- Open View > Subtitles (Ctrl+Shift+9 on Windows and Linux, Shift+Cmd+9 on a Mac).
- Click Detect speech…, choose the language and model in the Speech to Text dialog, and start it.
- Wait for two jobs in the Jobs panel: one exports the audio, the other converts it to text.
- The result appears in the Subtitles panel as a new subtitle track. Edit the text and the start and end times there.
Shotcut's docs list a known quirk: subtitle items "sometimes start earlier than expected". If the whole track runs consistently early, export it as SRT, shift it with a subtitle time shift tool, and bring it back with Import Subtitles From File. If only a few items are early, fix those times in the panel.
Shotcut: style and export with subtitles burned in
Shotcut's player doesn't show subtitles until you add the Subtitle Burn In video filter, and that filter only goes on the timeline's Output. From the filter's documentation:
- Select the timeline's Output, open Filters and add Subtitle Burn In.
- Choose the Subtitle Track to render.
- Set the font and size, text color, outline color and thickness, background color and padding, and opacity.
- Place it with Position X and Y; Horizontal Fit (left, center, right) and Vertical Fit (top, middle, bottom) set the alignment.
- Optional: the Typewriter group reveals text character by character, word by word or line by line.
- Export as usual. The subtitles become part of every frame.
When the export format supports it (MKV, MOV or MP4), Shotcut also embeds the subtitle track as a separate stream; tick Disable subtitles under Export > Other if you don't want that. The Subtitles panel exports SRT for platforms that take subtitle files.
Kdenlive vs Shotcut at a glance
| Docs read 19 October 2026 | Kdenlive 26.08 | Shotcut 26.9 |
|---|---|---|
| Speech engine | Whisper or VOSK | whisper.cpp (Whisper) |
| Setup | Python 3, about 2 GB of Whisper dependencies, then a model | Model downloader in the dialog |
| GPU | CUDA-capable GPU for Whisper | Vulkan on Windows and Linux, Metal on Apple Silicon |
| Language | Autodetect or chosen | Chosen in the dialog |
| Translation | To English with Whisper; others with SeamlessM4T | Not documented |
| Burn in | Default when rendering | Subtitle Burn In filter on Output |
| Subtitle file out | SRT or ASS | SRT, or embedded in MKV, MOV or MP4 |
| Subtitle file in | SRT, ASS, VTT, SBV | Import Subtitles From File |
Pick Kdenlive if you want a choice of two engines or built-in translation to English; pick Shotcut if you want the simplest setup, with no Python, and GPU speech recognition without an NVIDIA card. For how both compare with other desktop software and online tools, see desktop subtitle software vs an online generator, and for other open-source editors, the open-source video editors roundup. If you already have an SRT and only need it burned in, HandBrake, VLC or FFmpeg do it without an editor.
Styled 9:16 captions: the browser handoff
Kdenlive and Shotcut give you editable, plain subtitles. What they don't give you without a lot of keyframing is the short-video look: big animated words, the spoken word highlighted, one or two words at a time. For that, finish your vertical edit, render it without burned subtitles, and caption the MP4 in a browser.
CapsAI takes MP4 or MOV files of up to 10 minutes, transcribes them in one of 11 spoken languages (English plus 10 Indian languages), lets you fix every word and timing, and exports an MP4 with one of 37 caption styles burned in. The export keeps your video's shape, so a 9:16 render stays 9:16. It doesn't export or import SRT, so it can't feed Kdenlive's or Shotcut's subtitle track; it replaces that step for videos where the animated look matters more than a subtitle file.
What to do next
For subtitles on a long video, or a file for YouTube, use Kdenlive or Shotcut as above; both are free. For animated captions on a finished vertical clip, try the vertical video subtitle generator with the 3 free minutes, and browse the looks on the caption styles page. Packs start at $4 (₹49 in India, €4 in the euro area), paid once.
