In the Chrome Web StoreMP4 to Transcript help and support
Answers to the most common problems with the speech model, WebGPU, reading files and editing transcripts. If your problem is not here, email support.
First things to try
- Update Chrome to version 121 or newer and turn on graphics acceleration in Settings › System.
- Open
chrome://gpuand check that WebGPU says “Hardware accelerated”. - Keep the extension's tab open while a file is transcribed; you can use other tabs meanwhile.
- If the language was detected wrong, pick it under Wrong language? Transcribe again in.
Common problems and fixes
It says this browser can't run the speech model
Transcribing on your computer needs WebGPU, and Chrome reports that it isn't available. This happens on some older graphics chips, in virtual machines and when hardware acceleration is off. Turn on “Use graphics acceleration when available” in Chrome's Settings › System and restart Chrome, update Chrome and your graphics driver, then check chrome://gpu.
The model download stopped
Press Resume download: it continues where it stopped, also after you close the tab. “No connection to Hugging Face” means the network dropped or blocks huggingface.co; “Not enough free disk space” means the disk needs room for the model, at the size shown in the message: free some space and resume.
This file's audio can't be read
The file has no audio track, or its audio uses a format Chrome can't decode. If it plays in another app, convert it to MP3 or MP4 and try again. MP4, MP3, M4A, WAV, WEBM, MOV, MKV, OGG and FLAC usually work.
Transcription failed with a graphics error
The graphics chip stopped the speech model, usually because it ran out of memory. Close other heavy tabs or apps, such as games or video editors, and press Try again. If the graphics processor restarts during a file, the extension reloads the model and carries on by itself.
Transcription is slow
Speed depends on the graphics chip Chrome uses. On a laptop with two graphics chips, set Google Chrome to “High performance” in Windows Settings › System › Display › Graphics, then restart Chrome. Loading the model takes up to a minute each time the page opens.
I closed the tab before it finished
The text so far is kept. Open the transcript in the history, which is marked Unfinished, open the same recording again and press Resume. Only one file is transcribed at a time; you can start the next one when it's done.
The language is wrong or the text is in another language
Pick the right language under Wrong language? Transcribe again in above the transcript. That makes a new transcript and keeps the old one in the history. Accuracy is highest in widely spoken languages; for rarer ones, choosing the language by hand helps.
I can't play the recording next to an old transcript
After a restart, Chrome asks again for access to the file. Press Allow access, or Find the file… if it has moved. A transcript stays readable and exportable even without the file.
How do I free the disk space used by the model?
Open Settings and press Delete the model. Your transcripts stay; to transcribe again, the model is downloaded once more.
Your data
Your files are read on this computer and never uploaded. Transcripts and the history are kept in this browser's storage until you delete them; the speech model is kept in the browser's file storage. Delete a transcript with Delete transcript in the history, and the model with Delete the model in Settings. To keep a transcript elsewhere, export it as TXT, SRT or VTT. Removing the extension deletes the model and all transcripts.
When you write to us
- The file type and length, and the language you chose or that was detected.
- The graphics chip shown in Settings, and what
chrome://gpusays next to WebGPU. - The exact error message, and at what point (model download, loading or transcribing) it appeared.
- The extension name and its version from
chrome://extensions. - Your Chrome version from
chrome://settings/helpand your operating system. - What you did, what you expected and what happened instead.
- A screenshot, if the problem is visible on the screen.