Large-file workflow

How to transcribe a large audio or video file in your browser

A long recording asks more of your browser and device than a short voice memo. This guide helps you choose the right entry, prepare for the run, understand what happens between sections and avoid the interruptions that make you start over.

Short answer

Use standard transcription for a supported file up to 300 MB and 20 minutes. If either limit is too small, use the large-file tool for a local file up to 1 GB and 1 hour. Keep the tab open and the device awake; unfinished progress cannot be restored after a reload, closed tab or sleep.

01

Which transcription tool fits your recording?

File size alone does not tell you which workflow to use. Check both the media size and the audio-track duration. A highly compressed one-hour recording can be small in megabytes, while a short uncompressed WAV can be large. Whisper Web applies both limits.

Start with the standard tool when the file fits. It supports local files, microphone recordings and compatible direct media URLs. The large-file tool is for a selected local file and trades a longer limit for a more deliberate, section-by-section session.

Standard and large-file transcription limits in Whisper Web
What do you need?Standard transcriptionLarge-file transcription
Maximum file size300 MB1 GB
Maximum audio duration20 minutes1 hour
Input choicesLocal file, microphone or compatible direct URLLocal file only
How the recording is handledProcessed as a regular transcription jobDecoded and transcribed in five-minute sections
Can you pause between sections?No section-by-section pauseYes, after the current section finishes
Where does the completed result go?Local history in this browserThe same local history after the full recording completes
02

What should you check before starting a long transcription?

  • Confirm both limits

    Check the file size and recording duration before setting aside time. A file over 1 GB or an audio track over one hour must be shortened or split before Whisper Web can process it.

  • Choose a device you can leave awake

    A long local job uses the current CPU or GPU. A plugged-in laptop or desktop with clear ventilation is usually easier to leave running than a phone on battery.

  • Make sure the browser can decode the audio

    A familiar file extension does not guarantee that every browser supports the codec inside it. If analysis fails, export the source as MP3, WAV or M4A and try again.

  • Decide what you need at the end

    Use TXT for editable prose, JSON when another tool needs timestamps, or SRT/VTT for captions. Knowing the destination makes the review step more focused.

03

What does the browser do with a large recording?

  1. 1

    Inspect the selected media

    The browser checks the container, finds a readable audio track and measures its duration before transcription starts. This catches unsupported or over-limit files early.

  2. 2

    Plan five-minute sections

    Whisper Web divides the audio timeline into smaller working sections so it does not need to decode the entire recording into memory at once.

  3. 3

    Decode and transcribe in order

    Each section is decoded locally and passed to the selected Whisper model. Timestamped text is added in recording order before the next section begins.

  4. 4

    Save only after the full recording finishes

    When all sections complete, Whisper Web combines the text and timestamps and writes the transcript to local history. Until then, partial progress belongs to the current tab only.

04

How long will it take, and what will it do to your device?

There is no fixed turnaround time for a one-hour file. Processing speed changes with the selected Whisper model, WebAssembly or WebGPU backend, browser, processor, available memory, power mode and other work happening on the device.

The large-file page begins with a calibration message. After the first 30 seconds of audio are processed, it estimates a remaining-time range from the speed measured on your device. Treat that range as a planning aid, not a deadline: background tabs, thermal throttling and battery-saving modes can change the pace later.

Long inference can increase battery use, fan noise and device temperature. Connect to power when practical, keep ventilation clear and stop the job if the device becomes unusually hot. A smaller model usually asks less of the device, while a larger model downloads more data and uses more memory.

05

What happens if the transcription is paused or interrupted?

What Whisper Web keeps after common large-file interruptions
What happens?What is kept?What should you do next?
You request a pauseThe current five-minute section finishes; earlier sections remain in this tabResume with the next section when the device is ready
You refresh or close the tabUnfinished progress is lostSelect the source file and start again
The device sleepsUnfinished progress is lostKeep the device awake during the next run
The browser cannot decode the codecNo completed transcript is savedTry another browser or export the source as MP3, WAV or M4A
The full job completesText and timestamps are saved to local historyOpen the result to edit or export it
06

Does a large recording stay on your device?

Yes, when you choose a local file. Whisper Web reads, decodes and transcribes that file in the current browser rather than sending the media to a Whisper Web transcription server.

Local media processing does not mean the page makes no network requests. The site, fonts and selected Whisper model still need to download, and analytics can receive ordinary page and interaction data. Whisper Web does not send the selected media or transcript text to those analytics services.

After a successful run, the transcript is stored in IndexedDB in this browser. It stays there until you delete it or browser storage is cleared. Export a copy if the transcript matters; local browser history is convenient storage, not a backup plan.

07

What should you review before using a long transcript?

  1. 1

    Check names, numbers and specialist terms

    Automatic text is a draft. Return to the timestamp whenever a quotation, amount, date, negation or technical term matters.

  2. 2

    Inspect transitions between sections

    Read across section boundaries for repeated or missing words, especially when speech is fast, overlapping or cut by silence and music.

  3. 3

    Add structure the model does not create

    Whisper Web does not identify speakers or produce a summary. Add speaker names, headings and decisions yourself when the destination needs them.

  4. 4

    Export for the next tool

    Choose TXT or JSON for text work and SRT or VTT for timed captions. Review subtitle line breaks and timing in the destination player or editor.

08

Questions about this guide

Can Whisper Web transcribe a two-hour recording?

No. The large-file tool accepts an audio track up to one hour. Split a longer recording into shorter files before importing it.

Does the large file get uploaded?

No. A selected local file is decoded and transcribed on the current device. The site and selected Whisper model still use network downloads.

Can I close the browser and continue tomorrow?

No. Unfinished sections are kept only in the current tab. Keep the tab open and the device awake until the completed transcript is saved.

Do I need WebGPU for a large file?

No. WebAssembly is available for broader compatibility, while WebGPU is optional on supported browsers and hardware. Actual speed still depends on the device and model.

Which Whisper model should I choose?

Start with the default smaller model when device load and wait time matter most. Try a larger model when the device has enough memory and you can accept a larger download and longer processing time.

Can I edit and export the completed transcript?

Yes. Open the saved result in local history to edit it and export TXT, JSON, SRT or VTT.

Choose by file size and duration

Use the simplest workflow that fits your recording.