Whisper

"Faster Whisper" is a general-purpose speech recognition model. It is trained on a large dataset of diverse audio and is also a multi-task model that can perform multilingual speech recognition as well as speech translation and language identification.

Availability / Target HPC systems

• JupyterHub: best suited for interactive use (only for Tier3 HPC accounts)
• TinyGPU/Alex: best suited for processing of larger files in batch mode

Whisper on JupyterHub

Interactive usage of Whisper is available as a custom kernel in JupyterHub. The following steps are necessary:

1. Access JupyterHub according to this guide.

2. Select Whisper Transcription Web Service from the job profiles.

   

3. Choose whisper-demo from the available notebook types.

   

4. You can now use Whisper.

   

5. After you finished your work, remember to stop your instance manually by going back to the hub control panel (File > Hub Control Panel) and selecting Stop My Server. Closing the browser or logging out from JupyterHub will NOT free the resources!