Platform edition

Granite for Windows

IBM Granite Speech locally on the CPU or an NVIDIA GPU through a managed Python/PyTorch runtime.

Inside this edition

Under Processing device, choose Automatic, CPU, or NVIDIA CUDA and select Save settings; with CPU, the add-on downloads a smaller runtime without CUDA
Select Download model to install the runtime and the model, then select Use model
A later change of the processing device applies the next time the model loads; switching from CPU to Automatic or NVIDIA CUDA downloads the runtime again and keeps the model files
To free the disk space, select Remove model and runtime

Set up the add-on and verify its platform-specific capabilities with a first practical workflow.

Roles in TypeWhisper

Transcription

In TypeWhisper, open Settings > Discover plugins and install Granite, which is listed there as IBM Granite Speech (Local). The add-on then appears in the Settings sidebar, and its page opens. If it is already installed, open its page in the sidebar.
On the add-on's page, review the options available in this edition.
Under Processing device, choose Automatic, CPU, or NVIDIA CUDA and select Save settings; with CPU, the add-on downloads a smaller runtime without CUDA
Select Download model to install the runtime and the model, then select Use model
A later change of the processing device applies the next time the model loads; switching from CPU to Automatic or NVIDIA CUDA downloads the runtime again and keeps the model files
To free the disk space, select Remove model and runtime
Under Settings > Dictation, select Local (Granite Speech) as the Provider and choose a Model. A Dictation Only workflow can also use its own transcription model.

Transcription: Dictate a few sentences with Local (Granite Speech) and check the language, model, and result before you use it for longer recordings.