Microsoft AI for macOS

Availability

Microsoft AI 1.0.0 is available as a TypeWhisper for macOS plugin release. It requires TypeWhisper 1.7.0 and macOS 14 or newer. Microsoft currently marks MAI Transcribe in Azure Speech as a preview feature.

What the macOS add-on does

The add-on provides one cloud transcription engine: Microsoft AI (MAI Transcribe). When a microphone recording ends or an audio file is submitted, TypeWhisper normalizes it to WAV and sends it to the connected Azure Speech resource. The response can contain detected language, text segments, timing, and speaker metadata.

This edition does not stream partial text and does not support translation, LLM workflows, or text-to-speech. Its built-in catalog contains MAI Transcribe 2 and MAI Transcribe 1.5. A successful refresh adds compatible MAI models returned by Microsoft Foundry and caches them locally.

Settings

Setting	What it controls
Endpoint or resource name	Accepts the Azure Speech HTTPS endpoint without a path, or only the resource name. A bare name becomes https://<name>.cognitiveservices.azure.com.
API Key	Stores the Speech resource key in the macOS Keychain. The eye button reveals or masks the value while editing.
Save/Update connection	Normalizes and saves the endpoint and key. Saving a configured connection also attempts to refresh the model catalog.
Remove	Deletes the stored API key and makes the transcription provider unavailable until another key is saved.
Model	Selects MAI Transcribe 2 by default, MAI Transcribe 1.5 as a fallback, or another compatible discovered model.
Refresh icon	Requests the Foundry model catalog through the saved Azure endpoint. Built-in and cached models remain available if refreshing fails.
Transcript style	Clean removes fillers for readability. Verbatim preserves filler words, false starts, and self-corrections.
Speaker diarization	Requests speaker labels for MAI Transcribe 2. The control is disabled for MAI Transcribe 1.5.

The settings warn when a recognizable regional endpoint is outside East US, North Europe, Southeast Asia, or West US.

Setup tutorial

1. Update TypeWhisper to 1.7.0 or newer.
2. In the Azure portal, create a Microsoft Foundry resource for Speech in eastus, northeurope, southeastasia, or westus.
3. From the resource, copy the Speech endpoint or resource name and one API key.
4. Open TypeWhisper > Settings > Integrations.
5. Find Microsoft AI under Discover, install it, and make sure it is enabled.
6. Open its settings, enter the endpoint and API key, then select Save connection.
7. Keep MAI Transcribe 2 selected, choose a transcript style, and enable speaker diarization only if needed.
8. Select Microsoft AI as the transcription engine in a profile and make a short test recording.

Use MAI Transcribe

1. Select Microsoft AI as the transcription engine in the TypeWhisper profile that should use Azure.
2. Choose one input language when you know it with certainty. Leave language selection automatic for detection or code-switched audio.
3. Add product names, people, acronyms, and specialist vocabulary to the TypeWhisper dictionary. The plugin sends at most 500 terms and 20,000 total characters as phrase-list hints.
4. Use Clean for readable notes or captions; use Verbatim when the precise spoken form matters.
5. Enable speaker diarization before recording or importing multi-speaker audio. This setting requires MAI Transcribe 2.
6. Check the completed result after Azure returns it; no partial transcript appears while the recording is still running.

With MAI Transcribe 2, TypeWhisper requests segment timestamps. When Azure returns speaker information, the structured result keeps the label on each segment and the displayed text uses lines such as Speaker 1: ....

Audio limits

The macOS plugin accepts normalized WAV audio up to two hours and up to 300 MB. Larger input is rejected before the upload. Each Azure request has a three-minute network timeout; that timeout does not change the two-hour recording limit.

Troubleshooting

• The connection cannot be saved: Use an HTTPS Azure Speech endpoint without an extra path, query, user information, or fragment. You can also enter only the resource name.
• MAI Transcribe is unavailable in this region: Create or use a Speech resource in East US, North Europe, Southeast Asia, or West US.
• Foundry catalog unavailable: The connected Speech resource may not expose the model-catalog endpoint. MAI Transcribe 2 and 1.5 remain available locally.
• No speaker labels appear: Select MAI Transcribe 2, enable diarization before transcription, and confirm that Azure returned speaker metadata.
• Translation fails: The plugin intentionally rejects translate-to-English requests because MAI Transcribe does not support translation through this integration.
• Authentication fails: Recopy the key belonging to the same Azure Speech resource as the endpoint, then update the connection.

Cloud processing and privacy

Audio is sent to Microsoft Azure for transcription. The API key is stored in the macOS Keychain; endpoint, selected model, cached model IDs, style, and diarization preference are stored as local plugin settings. Review Microsoft’s current Azure privacy, preview, regional-availability, and pricing terms before submitting sensitive audio.

Sources: Microsoft MAI Transcribe documentation, Microsoft AI plugin source