A sample SwiftUI app using whisper.cpp to do voice-to-text transcriptions. See also: whisper.objc.

First whisper.cpp need to be built and a XCFramework needs to be created. This can be done by running the following script from the whisper.cpp project root:

$ ./build-xcframework.sh

Note: if you get the error "iphoneos is not an iOS SDK" then you probably need to run this command first:

sudo xcode-select -switch /Applications/Xcode.app/Contents/Developer

Open whisper.swiftui.xcodeproj project in Xcode and you should be able to build and run the app on a simulator or a real device.

To use the framework with a different project, the XCFramework can be added to the project by adding build-apple/whisper.xcframework by dragging and dropping it into the project navigator, or by manually selecting the framework in the "Frameworks, Libraries, and Embedded Content" section of the project settings.

1. Select a model from the whisper.cpp repository.¹
2. Add the model to whisper.swiftui.demo/Resources/models via Xcode.
3. Select a sample audio file (for example, jfk.wav).
4. Add the sample audio file to whisper.swiftui.demo/Resources/samples via Xcode.
5. Select the "Release" ² build configuration under "Run", then deploy and run to your device.

Note: Pay attention to the folder path: whisper.swiftui.demo/Resources/models is the appropriate directory to place resources whilst whisper.swiftui.demo/Models is related to actual code.

1. Follow all the steps in the Usage section, including adding the ggml model file.
2. Follow the Core ML support section of readme to convert the model.
3. Add the Core ML model (models/ggml-base.en-encoder.mlmodelc/) to whisper.swiftui.demo/Resources/models via Xcode.

When the example starts running you should now see that it is using the Core ML model:

whisper_init_state: loading Core ML model from '/Library/Developer/CoreSimulator/Devices/25E8C27D-0253-4281-AF17-C3F2A4D1D8F4/data/Containers/Bundle/Application/3ADA7D59-7B9C-43B4-A7E1-A87183FC546A/whisper.swiftui.app/models/ggml-base.en-encoder.mlmodelc'
whisper_init_state: first run on a device may take a while ...
whisper_init_state: Core ML model loaded