Whisper vs online transcription tool: which to pick

If you need to turn speech into text, you have two realistic routes: run Whisper, the open source speech recognition model, on your own computer, or upload your file to a web service. Neither is better in general. The right choice depends on how much audio you have, how comfortable you are with a terminal, how sensitive the recordings are, and what you want to do with the text afterwards.

What Whisper is, and what "Whisper-class" means

Whisper is an open source speech recognition model that you can download and run yourself. It transcribes speech in many languages and can detect the language on its own. Because the model is freely available, many tools are built on it or on models of similar quality.

That is what "Whisper-class" means: recognition of the same general level as Whisper, whether you run the original yourself or use a hosted service. Retranscrit, for example, uses Whisper-class speech recognition, with automatic punctuation and language detection across more than 90 languages. So the question is rarely "which engine is smarter?". It is "who handles the setup, the interface and the extras?"

Running Whisper locally: what you need

Local Whisper is free to use, but not free of effort. Plan for:

• Installation. You typically install Python packages and a media tool such as ffmpeg from the command line. Some community builds package this more simply, but you still manage updates yourself.
• Hardware. It runs on a CPU, but a recent GPU makes it much faster. On an older laptop, a long recording can take a long time to process.
• Model choice. Larger models are generally more accurate and slower. Picking one is your job.
• Output. You get text files and subtitle-style files with timestamps. Correcting mistakes means opening the output in another editor and matching it against the audio by hand.

If you already work in a terminal, this is a small hurdle. If you don't, expect an afternoon of setup before your first transcript.

Using an online tool: what you get without setup

An online tool removes the installation. You upload a file and get a transcript in the browser. With Retranscrit, that means:

• Upload formats: MP3, WAV, M4A, OGG, FLAC, AAC, MP4, MOV, WEBM and MKV.
• An editor with a synced player, clickable timestamps, search, inline correction and automatic saving.
• AI tools on top of the transcript: summary and key points, chapters, and on paid plans meeting minutes, a blog article draft, a LinkedIn post, translation that keeps timestamps, and a chat to ask questions about the content. Answers cite moments in the recording.
• An interface in English, French, Spanish, German and Arabic.

You can try it from the audio to text page, or from the video to text page if you start from a video file.

There are things it does not do. Retranscrit has no real-time transcription, no bot that joins Zoom, Teams or Meet calls, no mobile app, no browser extension, no public API and no integrations such as Zapier, Notion or Drive. If you need any of those, it is not the right tool.

Cost and volume: when local wins

Local Whisper costs nothing per minute. You pay with your time and your hardware. That makes it the better option when you have large volumes and don't want to pay for them.

Retranscrit's plans (prices include tax):

Plan	Price	Monthly AI transcription	Max file
Free	€0	30 min	15 min and 25 MB
Pro	€9.90/month or €99/year	600 min	4 h and 2 GB
Business	€24.90/month or €249/year	2,000 min	8 h and 5 GB

Pro includes a 7-day free trial (a card is required, and you can cancel at any time). Details are on the pricing page.

A rough rule: if you transcribe an occasional interview or a few meetings a month, an online plan is simple and inexpensive. If you process dozens of hours every week and don't need the editor or exports, a local Whisper setup will pay for itself in saved fees.

YouTube is a separate case. Retranscrit can turn a YouTube video into timestamped text in seconds, for free and without an account, but it reads the published subtitles (manual or automatic). If a video has none, you need to import the audio or video file instead.

Editing, timestamps, exports and speaker labels

This is where the two routes differ most.

• Editing. Local Whisper gives you raw output. A hosted editor lets you click a sentence, hear it, fix it and move on.
• Exports. Retranscrit exports TXT, SRT and VTT on every plan. PDF and DOCX come with Pro and Business, and JSON with Business. Local Whisper produces plain text and subtitle formats, and anything else is up to you.
• Speaker labels. Plain Whisper does not tell you who is speaking. Add-on tools exist, but they mean more setup. Retranscrit identifies speakers on Pro and Business only, not on Free.

If your end product is subtitles for a video, an SRT file from either route may be enough. If it is a clean interview transcript to share as a DOCX, the hosted route saves steps.

Privacy and data handling

Local Whisper wins on pure privacy: the audio never leaves your machine. For legally sensitive or confidential recordings, that is a strong argument.

With a hosted service, you should know exactly what happens to your files. For Retranscrit:

• Data is hosted in Paris, in the EU.
• There is no human review of your content.
• You can delete your data manually at any time.
• Without an account, the transcript and the file are deleted after 24 hours.
• With an account, the transcript text and AI results stay in your account. Only the imported audio or video file is deleted, after 7 days on Free, 30 days on Pro and 90 days on Business. YouTube transcriptions in an account are kept.

If your policy forbids uploading recordings to any third party, use local Whisper. If EU hosting and deletion controls are enough for you, a hosted tool is workable.

Accuracy: what to expect

Both routes rely on similar recognition quality, so results depend mostly on the recording. Retranscrit's site says accuracy is generally above 95% on clear audio. That is not a guarantee, and Whisper run locally has no guaranteed figure either.

What hurts accuracy, whatever the tool:

• Background noise or music
• Several people talking at once
• Distant microphones and echo
• Strong accents combined with poor audio
• Specialized vocabulary and names

Better audio helps more than switching tools. Record close to the speaker, and plan a quick proofreading pass for names and technical terms.

A simple decision guide

Choose local Whisper if:

• You have large volumes and don't want to pay for them.
• You are comfortable with the command line.
• Your recordings must stay on your own machine.
• Raw text or subtitle files are enough.

Choose an online tool like Retranscrit if:

• You want results without installing anything.
• You want to edit with a synced player and export to PDF or DOCX.
• You need speaker labels, summaries, chapters or translation (some on paid plans only).
• Your volume fits within 30 minutes a month (Free) or 600 minutes (Pro).

Choose neither if you need live captions, a bot in your meetings or an API. Retranscrit does not offer those.

Many people combine both: local Whisper for bulk batches, and a hosted tool for the few files that need editing and polished exports.

FAQ

Is Retranscrit the same as Whisper? No. Whisper is an open source model you run yourself. Retranscrit is a web service that uses Whisper-class speech recognition and adds an editor, exports and AI tools around it.

Is local Whisper really free? The software is free, but you supply the hardware, the setup time and the maintenance. Long files can be slow without a GPU.

Can I try an online transcription without creating an account? On Retranscrit, YouTube transcription from published subtitles works without an account, with a limit of 5 videos per day. Without an account, transcripts and files are deleted after 24 hours.

Which is better for confidential recordings? Local Whisper keeps files on your machine, which is the safest choice for highly sensitive material. A hosted service with EU hosting, no human review and manual deletion can be acceptable for less sensitive content.