Meta for Windows

Availability

Meta 1.0.1 is available as a TypeWhisper for Windows plugin release. Its current source code is maintained on the repository’s main branch.

What the Windows add-on does

Version 1.0.1 provides two capabilities through one Meta API key:

• Transcription: Muse Voice Transcribe handles live dictation and uploaded audio. It can include speaker labels when diarization is enabled.
• LLM workflows: Muse Spark processes TypeWhisper workflow prompts. The built-in fallback catalog contains Muse Spark 1.2 and 1.1.

Both capabilities use the same securely stored API key. The add-on loads compatible models from Meta’s /v1/models endpoint and keeps its saved catalog if a later refresh fails.

Settings

Setting	What it controls
API Key	Stores the Meta Model API credential in TypeWhisper’s secure secret storage. Test validates the key and refreshes the model catalog.
Transcription Model	Selects the Muse Voice Transcribe model. Muse Voice Transcribe 1.0 remains available as the documented fallback before a catalog is loaded.
Refresh	Reloads compatible muse-voice-transcribe-* and muse-spark-* models from Meta. Saved fallback models remain usable if refreshing fails.
Speaker diarization	Labels speakers in live and uploaded audio. Real-time transcription uses 16 kHz PCM over Meta’s WebSocket API.
LLM Model	Selects the Muse Spark model used by TypeWhisper workflows. Muse Spark 1.2 is the default fallback, with Muse Spark 1.1 also available.
Reasoning Effort	Controls Muse Spark’s internal reasoning with Minimal, Low, Medium, High, or X High. The default is Medium.

Changing the API key clears the previously fetched model catalog and normalizes both model selections. Test or refresh the connection to load the catalog for the new key.

Setup tutorial

1. Update to TypeWhisper 1.0.10 or newer and create an API key in the Meta Model API developer portal.
2. Open Settings > Integrations in TypeWhisper for Windows.
3. Find Meta under Discover, install it, and make sure it is enabled.
4. Open the add-on settings and paste the API key. It is saved as a protected secret.
5. Select Test. A valid key triggers a model refresh and reports how many transcription and language models were loaded.
6. Choose a Transcription Model and an LLM Model.
7. Enable Speaker diarization only when speaker-labelled output is useful.
8. Leave Reasoning Effort at Medium for a balanced starting point, or choose another level for Muse Spark workflows.

Use Muse Voice Transcribe

1. Open the TypeWhisper profile you want to use and select Meta as the transcription engine.
2. Set the expected input language or languages. The add-on forwards supported choices as language bias.
3. Add important names and specialist terms to the TypeWhisper dictionary. Version 1.0.1 sends them as keywords in both file and real-time transcription without disabling streaming.
4. Enable Speaker diarization before recording or importing a file when you need speaker labels.
5. Record a short test and check the final text before using the profile for longer audio.

In diarization mode, Meta’s completed turns are combined with speaker labels when available. When recording stops, TypeWhisper drains queued audio and waits for Meta’s terminal transcript before completing the result. Translation is not supported.

Use Muse Spark in a workflow

1. Create or edit a TypeWhisper workflow that contains an LLM processing step.
2. Select Meta as the LLM provider.
3. Choose the Muse Spark model configured in the add-on settings.
4. Start with Medium reasoning effort and test the workflow on representative dictated text.
5. Increase the reasoning effort for more involved instructions, or reduce it when faster, simpler processing is sufficient.

The Windows add-on supports parallel hedged LLM requests when TypeWhisper uses that execution mode.

Troubleshooting

• The model settings are hidden: Enter an API key first. The model section appears after a key has been stored.
• Invalid API key: Check the key in Meta’s developer portal, paste it again, and select Test.
• Refreshing models fails: The saved or built-in fallback catalog remains active. Retry Refresh after checking the connection and Meta API availability.
• No speaker labels appear: Confirm that Speaker diarization was enabled before the live session or file transcription and that Meta returned speaker information.
• A newly available model is missing: Use Refresh. Only model IDs beginning with muse-voice-transcribe- or muse-spark- are shown.

Cloud processing and privacy

Audio sent to Muse Voice Transcribe and text sent to Muse Spark are processed by Meta’s cloud API. Review Meta’s current terms, privacy information, regional availability, and pricing before using the add-on with sensitive content.