You want to practise speaking Spanish before a trip, but there is no conversation partner nearby. Your phone can give you a short practice conversation without an internet connection or a cloud speech service. OGAM (Off Grid AI Mobile) turns your speech into text and uses a local chat model for replies. OGAM Pro can also read those replies aloud with a downloaded voice. Prepare the models before you disconnect.
| Get OGAM for Android or iPhone | Current release |
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.
Use a small task, such as ordering lunch in Spanish or checking into a hotel in French. This gives the conversation a purpose and makes the language easier to review.
What do you need for offline language practice?
You need a multilingual transcription model and a local text model that can use your target language. For spoken replies, you also need Pro and a downloaded voice for that language. You can use free dictation and read the replies on screen if you do not need audio output.
The app supports Android 10 or later with at least 4 GB of RAM, and supported iPhones from iPhone 12 onward with iOS 17 or later. Model memory needs can exceed the app minimum. Complete installation, any Pro setup, and all required downloads while connected.
| Part | What to select |
|---|---|
| Speech recognition | A multilingual on-device model, such as Base marked 99 languages |
| Replies | A downloaded local text model that supports the target language |
| Optional spoken replies | An on-device voice and its language assets in Models > Voice |
Speech recognition and speech output have different language lists. The mobile voice runtime offers English, French, Spanish, Italian, Portuguese, Hindi, Polish, and German. The wider transcription list does not mean a spoken reply voice exists for every language. See the voice runtime documentation.
How do you set up a language practice session?
Start in Chat mode so you can check each transcript. Select the language you intend to speak, give the chat model a short role, and ask for one question at a time. Keep the replies short enough to read and answer without losing the thread.
- In Model Settings > Transcription (Speech to Text) > Transcription model, download and select a multilingual on-device model. Base is about 142 MB.
- In Chat Settings > SPEECH TO TEXT > Language, choose your target language.
- In Models > Text, select a downloaded local text model.
- Start a chat with a narrow practice request.
For example:
Practise beginner Spanish with me. You are a waiter taking my lunch order. Ask one short question at a time. After my answer, show one useful correction if needed, then continue. Use Spanish for the conversation and English for a brief correction. Do not invent a pronunciation score.
Use a fresh conversation when you change scenarios. A hotel check-in needs different words from a restaurant order.
How do you speak and check your answer?
Hold the app’s microphone control, say your reply, and release it. The recognized words appear in the message box. Compare them with what you meant, correct recognition errors, then tap Send. The chat model responds to that text.
Suppose you say the Spanish equivalent of, “I would like a table for two.” Check that the transcript retains the number. If it does not, fix the text before asking the model to correct your grammar.
This distinction matters: the chat model sees the recognized words. A mistaken transcript is not proof that your grammar or pronunciation was wrong.
If a correction is unclear, ask the model to show both sentences and explain the change briefly. AI corrections can also be wrong.
How do you hear the reply aloud?
With Pro, open Models > Voice and download the local voice model. In Chat Settings > TEXT TO SPEECH, enable Enable TTS in Chat mode. Select a supported Language and Voice. Wait for the selected voice assets to be ready before going offline.
Long-press an assistant message and choose Speak. Use Speed to reduce playback speed if needed. Selecting a Spanish voice does not translate an English answer; ask the text model to write the answer in Spanish.
Stay in Chat mode for your first practice session. Checking the recognized words before sending makes it easier to separate a speech recognition error from a language mistake.
Can it assess your pronunciation?
This is conversation practice, not a validated pronunciation score or accent assessment. A correct transcript does not prove correct pronunciation. Check unfamiliar advice with a trusted learning reference or teacher.
What should you check when practice stalls?
| Problem | Likely check | Next action |
|---|---|---|
| The transcript appears in English | Speech model and Language | Select a multilingual model and the intended language. |
| The answer is too difficult | Practice request | Ask for shorter sentences and one question at a time. |
| The model corrects words you did not say | Transcript | Correct recognition errors before sending. |
| Replies stay silent | Voice readiness and Pro | Finish the voice download, enable TTS, and check device volume. |
| A stage needs internet | Active model selections | Check speech, chat, and voice separately for remote selections. |
How do you check that the session works offline?
After setup, turn on airplane mode and switch Wi-Fi off. Ask one short question, dictate a reply, and play an answer if you use TTS. This checks the parts you actually plan to use. Keep the app open for the session; this guide does not promise listening while the phone is locked.
With local models selected, those stages run on your phone. External tools and configured device sync are separate. Start with a simple role-play that needs no web lookup.
Download OGAM, prepare the local models, and try a five-turn restaurant conversation with internet access off. Keep one useful phrase and practise saying it again.