You can listen to AI replies in supported languages on your Mac without using a cloud speech service. OGAD (Off Grid AI Desktop) generates the voice locally. Select a language and voice in Settings > Voice, let the required files download, then use Speak on an assistant reply. Playback can run offline after setup.
| Download OGAD for Mac | 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.
The desktop speech-output workflow is part of the free core app. You do not need OGAD Pro for the settings and message playback described here.
This is useful for listening to a draft, checking how a translated reply sounds, or hearing an answer while looking away from the screen. It reads assistant messages inside OGAD.
What do you need for local speech on Mac?
Use the current OGAD build on an Apple Silicon Mac. The desktop download page lists M1 and later Macs. Keep internet access for the initial app and voice-asset downloads, and allow enough free storage for those files.
You also need:
- A working speaker or headphones.
- A local voice selected in the app.
- A completed assistant message to read aloud.
- A downloaded local text model if you want new AI replies without internet.
Speech generation and chat generation are separate. You can test playback on an existing reply before asking the chat model to produce more text.
The desktop feature list includes text-to-speech in the free app. The mobile app uses a different feature tier, so do not apply mobile Pro requirements to this Mac workflow.
Which languages and voices are available?
The current local desktop voice runtime provides US and UK English, French, Spanish, Italian, Portuguese, Hindi, Polish, and German options. Settings > Voice > Language shows the choices available from the active runtime. After choosing a language, the Voice menu shows matching speakers.
Some examples from the current local catalog:
| Language option | Example voice |
|---|---|
| English (US) | Heart |
| English (UK) | Emma |
| French | Siwis |
| Spanish | Dora |
| Italian | Sara |
| Portuguese | Dora |
| Hindi | Alpha |
| Polish | Mateusz |
| German | Anna |
Some voice names repeat across languages. Select Language first so you choose the speaker from the intended group.
Use the actual menu as the source for your installed build. A language mentioned in general model documentation does not necessarily have a voice in the app’s installed runtime.
How do you prepare a voice on your Mac?
Open Settings > Voice, choose Language, then Voice. The app prepares the selected voice and downloads missing assets. Wait for the language’s voice ready message before testing offline. A different language may need additional files even if you have already used an English voice.
1. Check that the voice model is local
If you previously selected a remote voice service, return to Models > Voice and select the downloaded local voice model with Use.
Local and remote speech services can both appear in the desktop app. A working voice while connected is not enough to establish offline processing.
2. Choose Chat mode and enable speech
In Settings > Voice, set Interface mode to Chat. Set Text-to-speech to On.
This keeps replies as text with optional playback. Voice conversation mode is a separate interaction; you do not need it for this first test.
3. Choose the language and speaker
Select a language, then a voice. For example, choose French, then Siwis.
Wait while the app checks or downloads the voice assets. Continue when it says the selected language’s voice is ready. Leave Playback speed at its default for the first check.
4. Check the audio output
Select Test voice. The built-in test speaks an English sample sentence. It checks that sound can be generated and played; it is not a full pronunciation test for the selected language.
For a language check, use an assistant reply written in that language, as described below.
How do you read a reply in another language?
Generate or open a completed assistant reply written in the language you selected. Use its Speak action to hear the message. The voice reads the text provided to it; choosing a French or Hindi voice does not translate an English reply into that language.
For a French test, ask a local chat model:
Reply in French with two short sentences explaining how to prepare for a train journey.
Read the reply first. Check that it is actually in French and contains the information you asked for. Then use Speak on that assistant message.
Use a short passage first. Names, abbreviations, and numbers are useful details to check when you compare voices. Speech output can have pronunciation errors even when the written reply is correct.
If you want to listen to your own text, ask the local chat model to repeat a short passage exactly. Compare the assistant’s output with your original before playing it. A chat model can alter text, so this is not a guaranteed verbatim document-reading workflow.
How do you confirm that speech works without internet?
Prepare the voice while connected, then turn off Wi-Fi and disconnect Ethernet. Open an existing completed assistant reply and choose Speak. That tests local text-to-speech separately from the model that generated the reply.
If you also want to test a new offline answer, select a downloaded local chat model before sending another prompt. Then play that answer with the prepared local voice.
Repeat the check for each language you plan to use offline. Downloading one voice does not prove that all language assets are available on your Mac.
What should you check if playback fails?
| Problem | Check | Action |
|---|---|---|
| Speak is missing from a reply | Settings > Voice | Use Chat mode and set Text-to-speech to On. Wait for the reply to complete. |
| A language fails only while offline | Voice assets may be incomplete | Reconnect, select that voice, and wait for its ready message before disconnecting again. |
| Test voice runs but there is no sound | macOS output device and volume | Check the selected speaker or headphones and the system output level. |
| The voice sounds wrong for the text | Language and Voice | Choose a voice that matches the text’s language. |
| Changing Language did not translate the text | The written reply | Translate the text in a separate chat step, review it, then play it. |
| A new AI reply fails offline | The chat model may be remote | Activate a downloaded local text model. |
| The app reports a missing voice runtime | The installed desktop build | Update or reinstall the official Mac build. A model file alone does not replace the runtime. |
| Voice preparation fails | Download completion and free resources | Finish setup while connected, close memory-heavy applications, and try a short passage. |
If a voice has trouble with a long answer, start with a shorter reply. Do not use a successful sentence as evidence that a whole document will be read accurately.
Does the text go to a speech server?
With the local voice selected, the Mac processes the text using the local speech runtime and produces audio on the device. This workflow does not require a cloud speech API key.
A selected remote voice service changes where speech is generated. A remote chat model can also receive the prompt before speech playback begins. Check both selections if you want the full conversation to remain local.
Local synthesis creates audio for playback. It does not mean that no temporary files exist on your Mac. Sharing the message or copying it into another service is a separate action.
Mac text-to-speech FAQ
Is multilingual speech output a Pro feature on Mac?
No. Text-to-speech in the desktop chat is part of the free core. This differs from the mobile app’s Pro speech-output tier.
Does a voice translate a message?
No. It speaks the text it receives. Use the chat model for translation, review the result, then choose a matching voice.
Does Test voice test the selected language?
The current test uses an English sentence. Use a reply written in the target language to assess that voice’s pronunciation.
Can I install these voices as macOS system voices?
These steps configure OGAD. They do not install a system-wide voice for other Mac applications.
Do I need a microphone?
No microphone is needed to play a typed assistant reply. A microphone is only needed if you also want to speak your requests.
Choose one language, wait for its voice-ready message, and listen to a short reply with your Mac disconnected from the network.