You can turn spoken French into editable text on your Mac or Windows PC without uploading the audio. OGAD (Off Grid AI Desktop) runs a downloaded speech model on your computer. Select a multilingual model, choose French, and record through the app’s microphone. After setup, this works without internet.
| Download OGAD | Desktop releases |
What would you like to do with Off Grid AI?
Have a feature or use case you would like us to support? Tell us what you want to do and which device you use.
Write to support@offgridmobileai.co, join our Slack community, or talk to us on Reddit.
This is a guide to French dictation inside the app. You can speak a draft email or a note, check the words, and copy the result. The microphone workflow gives you text to review before you send it to an AI model.
Saved audio files use a separate project import workflow. There is a short explanation of that option below.
What do you need to transcribe French offline?
You need OGAD, a microphone, and a downloaded multilingual transcription model. The desktop download page provides builds for Apple Silicon Macs and Windows x64 PCs. Download the app and model while connected, then use local transcription without a network connection.
Before recording, check these items:
- Your microphone works and the app has permission to use it.
- The speech model download is complete.
- You have enough free storage for the model and working files.
- Other applications leave enough memory available for the model to load.
You do not need an English keyboard layout, a French operating system, or a cloud speech API key. The spoken-language setting is inside OGAD.
A chat model is optional for this first task. The speech model returns the transcript. A local chat model is needed if you also want to summarize, translate, or rewrite that text while offline.
Which model should you use for French speech?
Choose a multilingual Whisper model. Whisper Base is a small starting point for a short recording. If it misses too many words, compare it with Whisper Small using similar audio. Avoid English-only model files with .en in the name; they do not add French support when you change a setting.
| Desktop model | Approximate download | Use it to |
|---|---|---|
| Whisper Tiny | 78 MB | Check the recording workflow with a small download |
| Whisper Base | 148 MB | Try French dictation with your own microphone |
| Whisper Small | 488 MB | Compare recognition when Base misses words |
| Whisper Medium | 1.53 GB | Test a larger model if your computer has enough memory |
These are rounded decimal file sizes from the desktop catalog. They are not RAM requirements. The size shown in the app can differ with the file variant or unit conversion.
For example, ggml-base.bin is multilingual. ggml-base.en.bin is English-only. The Whisper engine documentation describes these model variants.
There is no accuracy guarantee for a particular French accent. Your microphone, room noise, speaking pace, and model all affect the result. Use a recording you can check before you rely on the transcript.
How do you select French in OGAD?
Open Models > Transcription, download a multilingual Whisper model, and select Use. Then open Settings > Transcription. Check Current model and choose French under Spoken language. This sends French as the language setting for the next transcription.
1. Download the model
Find Whisper Base in Models > Transcription and select Download. Wait for the download to finish.
Choose the multilingual file if the app offers variants. The filename must not contain .en for this workflow.
2. Set it as active
Select Use on the downloaded model. Its card should show Active.
In Settings > Transcription, confirm that Current model is the downloaded local model. A model can be stored on your computer without being selected for the next recording.
3. Set Spoken language to French
Choose French under Spoken language. This is useful for a short French note where automatic language detection has little context.
You can use Auto-detect with a multilingual Whisper model later. If the selector only offers English, check the model selection first. Choose multilingual Whisper instead of an English-only model.
How do you record French and review the text?
In the normal text chat screen, click Record voice, speak, and click Stop recording. The transcript appears in the message box. OGAD lets you edit that draft before sending it. Stay in text chat for this test; voice conversation mode can send a transcribed turn automatically.
After setup, turn off Wi-Fi and disconnect Ethernet if connected. On a Mac, make sure any wired adapter is disconnected too.
Use the microphone control inside OGAD. Your operating system’s dictation shortcut uses a different workflow.
For a first recording, try this sentence:
Le rendez-vous avec Madame Dupont est vendredi à quinze heures, au vingt-deux rue Victor-Hugo.
Check these details in the result:
- Madame Dupont: Is the name correct?
- Vendredi: Did the day stay the same?
- Quinze heures: Did the model preserve the time?
- Vingt-deux rue Victor-Hugo: Are the number and street name correct?
The model may write a number as digits or words. Check the meaning before you change the format.
Also review accents and apostrophes. French pairs such as a / à and ou / où can change meaning. A readable sentence is not proof that every word is right.
Correct the draft and copy it into your document or email. If you want an AI reply while offline, select a downloaded local chat model before sending the text.
Can you use an existing French audio file?
OGAD also transcribes audio added to a project’s knowledge base. That is a separate core feature. It prepares the recording as source material for project questions. It does not put the transcript into the same editable message box used by microphone dictation.
Set the local transcription model and Spoken language > French first. In your project, Knowledge base > Add files accepts audio. The desktop feature list includes audio uploads in Projects.
Use project import when you want to ask questions about a saved recording. Keep the project’s AI models local for an offline workflow. This article’s numbered steps cover microphone dictation, not a standalone transcript export or subtitle workflow.
What should you check when French transcription is wrong?
Start with the selected model and spoken language. An English-only model cannot handle this setup as described. With a multilingual model active, select French explicitly, reduce background noise, and try a complete sentence before downloading a larger model.
| Problem | What to check | What to do |
|---|---|---|
| French is missing from the selector | The active speech model may only support English in this workflow | Select multilingual Whisper Base or Small, then check Spoken language again. |
| Output appears in another language | Automatic detection may have too little audio | Set Spoken language to French and record a full sentence. |
| Names or addresses are wrong | The spoken name may be unfamiliar or unclear | Correct it in the draft. Try a quieter recording and compare another model if needed. |
| No audio records | Microphone permission or selected input | Allow the app to use the microphone and check the system input device. |
| The model does not load | Free memory and download completion | Close memory-heavy apps, finish the download, or choose a smaller model. |
| Transcription fails when disconnected | A remote transcription service may be selected | Select the downloaded local model again and check Current model. |
| Text appears, but no AI answer follows | The active chat model may be remote | Choose a downloaded local chat model for an offline reply. |
Avoid asking the speech model to handle a long, noisy recording as your first test. A short sentence makes it easier to identify a microphone problem, a model choice problem, or a language setting problem.
Does French transcription upload your audio?
The local transcription path processes the audio on your computer with the selected speech model. No cloud speech API is required. The app uses local working files during transcription, so local processing does not mean that no files are created.
A remote transcription service changes where audio is processed. Sending the transcript to a remote chat model is a separate network action. Check both model selections if your full workflow must stay local.
French transcription FAQ
Is this free on Mac and Windows?
In-app speech input is part of the free desktop core. You do not need Pro for the microphone steps in this article. Dictation directly into other applications is a separate feature and is outside this workflow.
Can it recognize Canadian or other regional French accents?
The selector offers French, rather than a separate setting for each accent. Try a short recording with your own voice. Compare results and correct the text; this guide does not claim equal accuracy for all regional accents.
Does it translate French into English automatically?
The workflow here transcribes French speech as French text. For English, review the French transcript and ask a local chat model to translate it. That is a separate step with a separate model.
Can I use French and English in the same sentence?
You can test mixed-language recordings, but multilingual support does not promise accurate handling of every switch. Start with French-only speech, then compare with the mixed-language notes you actually need.
Install OGAD, select French, and record one sentence that includes a name and a time. Check the text before you use it.