PotPlayer Whisper: How to Generate Subtitles From Audio
PotPlayer Whisper turns a video's audio into subtitles on Windows. Which engine and model to pick, how real-time mode works, where the SRT goes and what fails.
PotPlayer Whisper support turns the audio of a video into subtitles inside the player. Right-click the video, open Subtitles, choose Create subtitles from audio..., pick an engine, a model and the language, and start. PotPlayer downloads what it needs on first use and can save the result as an .srt next to the video.
The feature arrived in PotPlayer's 241211 build, released on 11 December 2024, and has grown since: more engines, an update button, VAD and a real-time mode for streams. Below: the steps, which engine and model suit your PC, real-time mode, translation, the usual failures and where the feature stops. Facts come from PotPlayer's official version history, its English interface and its engine developers' own repositories, checked on 3 November 2026.
How do you generate subtitles with Whisper in PotPlayer?
Six steps. The menu and setting names are the ones PotPlayer's English interface uses.
- Open a video file. The regular mode works on files only; PotPlayer answers a stream with "Only files are available."
- Right-click the video → Subtitles → Create subtitles from audio...
- Choose the Whisper AI engine. The first time, PotPlayer asks "Does download the engine?"
- Choose the Whisper AI model and the Whisper AI language. The model download is offered the same way.
- Tick "If possible, save subtitles in the same folder as the video". The 250226 build added saving "with the same name as the video", so you get
movie.srtbesidemovie.mp4. - Start. PotPlayer extracts the audio, runs the engine and loads the subtitles.
Two more options sit in the same window. Automatically generated when playback starts runs the job whenever a file opens, which suits a folder of lectures more than a film you watch once. Applying voice emphasis filter when extracting sound emphasises the voice in the extracted audio, and is worth a try when music runs under the speech.
PotPlayer is free and runs on Windows 11, 10, 8.1, 8 and 7, according to potplayer.tv. The current build is 261021.
Which Whisper engine should you pick in PotPlayer?
Pick by graphics card. The Whisper AI engine list offers seven engines, all running OpenAI's Whisper models in different builds. The download sizes are the ones on Kakao's server on 3 November 2026:
| Engine | What it is | Download | Use it when |
|---|---|---|---|
| CPU | whisper.cpp on the processor | 3.9 MB | No usable GPU |
| BLAS | whisper.cpp with BLAS math on the processor | 23 MB | No GPU, and CPU is too slow |
| Vulkan | whisper.cpp on the GPU through Vulkan | 35 MB | AMD or Intel graphics |
| CUDA | whisper.cpp on an NVIDIA GPU, CUDA 12 runtime included | 617 MB | NVIDIA, smaller download |
| Const-me | A Windows port of whisper.cpp on the GPU through Direct3D | 0.4 MB | A tiny GPU engine without CUDA |
| Whisper-Faster | Purfview's older faster-whisper build | 894 MB | Only if an older setup needs it |
| Faster-Whisper-XXL | Purfview's current faster-whisper build | 2.5 GB | NVIDIA, the most options |
CPU, BLAS, Vulkan and CUDA are builds of whisper.cpp. Const-me's Whisper describes itself as "a Windows port of the whisper.cpp implementation". The two Faster-Whisper engines come from Purfview's whisper-standalone-win, whose README lists PotPlayer among the programs it is meant for and says the executables "automatically will choose to work on GPU if CUDA is detected". Its author marks the older Faster-Whisper build as deprecated, which is why Faster-Whisper-XXL is the one to use on NVIDIA.
Faster-Whisper-XXL arrived in the 250226 build, together with an Update button for the engines. Updating does not work while a video plays: PotPlayer asks you to stop playback first.
PotPlayer Whisper models: which one is right?
The largest model your PC finishes in a time you accept. The whisper.cpp engines use these model files, again with the sizes on Kakao's server:
| Model | Download | Notes |
|---|---|---|
| tiny | 74 MB | Fastest, for a rough idea of what is said |
| base | 141 MB | Still rough on accents and noise |
| small | 465 MB | The realistic minimum on a CPU |
| medium | 1.46 GB | Superseded by turbo: slower, rarely better |
| large-v3-turbo | 1.55 GB | The sensible default on a GPU |
| large-v3 | 2.95 GB | Slowest; not always better than turbo |
English-only .en variants exist for the smaller models and do better when every word is English. The two Faster-Whisper engines use faster-whisper's own model format rather than these files.
Bigger is not automatically better. In our Whisper model comparison, large-v3 skipped a stretch of speech that a smaller model caught on the same clip. That is one clip, not a benchmark, but it is a reason to start with large-v3-turbo rather than the largest file.
The 250625 build added VAD, voice activity detection, which skips silence and reduces the invented lines Whisper writes over music and pauses. Those invented lines are covered in why Whisper repeats lines and invents text.
Real-time mode: PotPlayer subtitles while you watch
Since build 250909, PotPlayer can also Create subtitles from audio(Real-time): it transcribes as playback runs, with three settings, Not used, Use if there are no subtitles and Always use. One more option limits it to sources that are not files: "Use only when generating subtitles from sound is not possible(not a file)".
So the two modes split the work. A file on your disk goes through the regular mode, which processes the whole audio and can save an .srt. A live stream or other non-file source goes through real-time mode, which subtitles what is playing now. The 260114 build fixed real-time subtitles drifting out of sync during live streams, so update before you judge it.
Can PotPlayer translate Whisper subtitles?
Yes, through its subtitle translation, which is older than the Whisper feature. In the Subtitle Translation menu you choose a Translation Engine and whether to show the translation only, above the original or below it. The current installer ships translation plug-ins for Google, Bing, DeepL, Papago, Yandex, LibreTranslate and LM Studio; DeepL and Papago ask for your own API key.
A newer Builtin AI translation option runs a local language model, which PotPlayer downloads in the GGUF format, so the text stays on your PC. The online engines send the subtitle text to their service, which matters for confidential video.
The 220706 build added "the ability to save translation when saving subtitles", so a translation can end up in the saved file. For a translation to keep long-term, translating a corrected .srt is the safer route; translating a video into your language compares the options.
PotPlayer Whisper not working: what to check
Work down this list:
- Update PotPlayer. Whisper has changed in nearly every build since December 2024. The 250313 build fixed Whisper-Faster on NVIDIA 5000 series cards, and the 250514 build improved how extra Whisper options are handled.
- The engine or model did not download. Use Open engine folder and Open model folder to see what arrived. PotPlayer's developer has said on the Korean user forum that the files can be downloaded by hand and placed there when the automatic download fails.
- Non-English characters in your Windows user name. In a reply on the same forum, the developer traced one failure to Korean characters in the account name and suggested an account without them. If your user name contains Chinese, Japanese, Cyrillic or accented letters, that is worth ruling out too.
- Out of GPU memory, or far too slow. Choose a smaller model, or switch from CUDA to Faster-Whisper-XXL or the reverse.
- It refuses a stream. "Only files are available" means the regular mode; use real-time mode for that source.
If the text comes out but is wrong, check the language setting first. PotPlayer's own notice is blunt about where its responsibility ends: "PotPlayer is not involved in Whisper AI conversion", and for an abnormal result it asks you to "contact the engine developer".
Where does PotPlayer's Whisper fall short?
When the subtitles are meant to be published rather than read while watching. Purfview, who builds the Faster-Whisper engines PotPlayer runs, said it directly in a discussion on his repository: "PotPlayer is not the right tool to generate proper subtitles. As far as I know, it takes text from the progress output, which isn't actual subtitles."
PotPlayer is a player. Its Browse/Edit/Sync Subtitles window (Alt+E) fixes a line or shifts the timing, but there is no waveform timeline, no reading-speed check and no caption presets for a finished video. For understanding a foreign video tonight, that is fine. For YouTube, a client or a course, the file needs an editor afterwards.
When does Sablate fit better than PotPlayer?
When you want subtitles to deliver: corrected, styled and exported, on Windows or Mac.
Sablate is a desktop app for Windows 10 or later and macOS 12 or later. It transcribes with faster-whisper in four tiers, Fast (tiny), Balanced (small), Accurate (large-v3-turbo) and Best (large-v3), with no engine to choose. On Windows with an NVIDIA GPU, the CUDA runtime downloads on first GPU use. Its editor retimes cues on a timeline, splits and merges them and warns when a line is too fast to read.
One transcription then feeds several translation layers, each with its own style and font for its script. Thirteen caption presets, a word-level karaoke sweep and burn-in at 9:16, 1:1 or 16:9 turn the result into a finished video.
The free plan needs no account and covers videos up to ten minutes, three jobs a month, the Fast and Balanced models and one offline translation layer; its burn-in carries a small corner mark. Pro costs $29 once or $5.99 a month; the pricing page lists the rest.
PotPlayer vs Sablate at a glance
| PotPlayer | Sablate | |
|---|---|---|
| Price | Free | Free plan; Pro $29 one-time or $5.99 a month |
| Platforms | Windows | Windows and macOS |
| What it is | A video player | A subtitle app |
| Speech engines | 7 Whisper builds, chosen by you | faster-whisper, four tiers |
| Subtitles while a stream plays | Yes, real-time mode | No |
| Saved output | SRT next to the video | SRT, VTT, TXT on Free; ASS and multi-track MKV on Pro |
| Editing | Browse/Edit/Sync window | Timeline editor with reading-speed warnings |
| Translation | Online engines and a built-in local AI option | Layers on one transcription; Argos offline, or DeepL, OpenAI or Claude with your key (Pro) |
| Caption presets for a finished video | No | 13 presets, burn-in at 9:16, 1:1 or 16:9 |
PotPlayer wins on price, real-time subtitles for streams and the choice of engines and translators. If you mainly want to follow a video in a language you do not speak, stay with it. Sablate is the route when the subtitles have to leave your PC looking finished.
Which route should you take?
"I want to understand a video I'm watching." PotPlayer with large-v3-turbo, and its subtitle translation if the language is foreign.
"I have an NVIDIA card." Faster-Whisper-XXL or CUDA with large-v3-turbo.
"I have AMD or Intel graphics." Vulkan with small or large-v3-turbo, depending on how long you will wait.
"I'm on a Mac." PotPlayer does not run there. Subtitle Edit with Whisper is the free route; a desktop subtitle app is the shorter one.
"I need subtitles for YouTube or a client." Generate them in an editor, not a player, so the timing and line lengths get checked.
"Styled captions in three languages for one video." Download Sablate for Windows or Mac.
Frequently asked questions
How do I use Whisper in PotPlayer?
Open a video file, right-click it, open the Subtitles menu and choose Create subtitles from audio. Pick a Whisper AI engine, a model and the spoken language, then start. PotPlayer asks to download the engine and the model the first time, extracts the audio, runs Whisper and loads the subtitles when it finishes.
Which Whisper engine should I choose in PotPlayer?
On an NVIDIA graphics card, Faster-Whisper-XXL or the CUDA engine; on an AMD or Intel GPU, Vulkan; with no usable GPU, CPU or BLAS. Faster-Whisper-XXL is the largest download at about 2.5 GB, while the CPU engine is under 4 MB. The older Whisper-Faster engine has been deprecated by its author.
Where does PotPlayer save Whisper subtitles?
Next to the video, as an SRT file with the same name, when the option to save subtitles in the same folder as the video is on. PotPlayer added saving with the video's name in its 250226 build. Tick the option before you start, so the result lands as a file you can open in any subtitle editor or player.
Can PotPlayer generate subtitles in real time?
Yes. Since build 250909, PotPlayer can generate subtitles from audio in real time while something plays. You can set it to never run, run only when there are no subtitles, or always run. It is meant for sources that are not files, such as live streams; for a file on your disk, the regular mode processes the whole file.
Why is PotPlayer Whisper not working?
Usually a failed engine or model download, an old PotPlayer build, or a graphics card the engine does not support. Update PotPlayer first, since builds in 2025 fixed NVIDIA 5000 series cards. If a download fails, place the files by hand through Open engine folder and Open model folder. The developer has also traced one failure to Korean characters in the Windows user name.
Is PotPlayer available for Mac?
No. PotPlayer's official site lists Windows 11, 10, 8.1, 8 and 7 only, in x86, x64 and ARM64 builds. On a Mac you need another tool to generate subtitles from audio, such as a desktop subtitle app or Subtitle Edit, both of which run Whisper locally on macOS.