| Both sides previous revision Previous revision Next revision | Previous revision |
| dcc:itsol:whisper:setup [2025/10/29 14:28] – added header giulio | dcc:itsol:whisper:setup [2026/09/25 10:19] (current) – alba |
|---|
| {{indexmenu_n>1}} | {{indexmenu_n>1}} |
| ===== Initial Setup ===== | |
| |
| ==== Setting up on HPC using the portal ==== | ====== Before you start ====== |
| |
| The guide below will take you through some basic steps to start running automatic transcription jobs on the High Performance Computing (HPC) cluster from a Windows computer. Before you can get started with setting Whisper up, you need to first [[https://portal.hb.hpc.rug.nl/public/start.html|Request a Hábrók (HPC) account or log into the HPC portal]] | Whisper runs on Hábrók and reads your recordings from one directory there, writing the transcripts to another. You need: |
| |
| {{ :dcc:itsol:whisper:hb_portal_1.png?direct&900 | }} | * a Hábrók account; |
| | * space in your Hábrók ''%%scratch%%'' directory for the recordings and the results; and |
| | * the language spoken in the recording, if you know it. |
| |
| Once you have an HPC account, you can navigate to the ''Files'' tab in the top menu bar and select ''/scratch/p-number''. Before we can run the script for the transcription, we need to make sure that folders are set up correctly in your HPC environment. | ===== Get access ===== |
| |
| {{ :dcc:itsol:whisper:hb_portal_2a.png?direct&900 | }} | **Request an account.** Staff members and PhD students request a P-account. Students request an S-account, which needs approval from their course, thesis, or project supervisor. Both request forms are on the [[https://portal.hb.hpc.rug.nl/public/start.html|Hábrók web portal start page]], and [[:habrok:introduction:policies|Accounts, access and policies]] explains who is eligible. You can skip this step if you already use Hábrók. |
| |
| In this window, you will be able to manage your files and folders in HPC. To begin with, make sure that you are in the folder with your p-number in the ''/scratch/'' partition of Hábrók. In the next steps, you will need to create a main folder for Whisper, then two subfolders for the input and output data. To create the main folder, click on ''New Directory'' as shown below. Please make sure that you write "whisper" in lower case. When the folder appears in your window, click on its name to open the next window. | **Log in to the web portal.** Once your account is active, open the [[https://portal.hb.hpc.rug.nl/public/start.html|start page]] and use the Login link in the top right corner. Everything in this guide happens in the portal. |
| |
| {{ :dcc:itsol:whisper:hb_portal_3.png?direct&900 | }} | {{ :dcc:itsol:whisper:login.webm?800x450 |}} |
| |
| Once inside the Whisper main folder, you need to create two subfolders. Once again, use ''New Directory'' to create each new folder. Call them "input" and "output", respectively, taking care to use lower case letters. Do not worry if your do not see any ''.sh'' file or ''slurm'' file in your file view. They will come later. | ===== Check your scratch space ===== |
| |
| {{ :dcc:itsol:whisper:hb_portal_4.png?direct&900 | }} | The app reads recordings from and writes results to your ''%%scratch%%'' directory. The default quota is 250 GB. ''%%scratch%%'' is not backed up, so keep the results you need somewhere else. See [[:habrok:data_management:storage_areas|Storage areas]] for details. |
| |
| ==== Upload your data before launching the job ==== | ===== Create input and output directories ===== |
| |
| Now that you have the folder structure ready, you can upload the audio file(s) you wish to transcribe. Click on the "input" folder to open its window view, then click ''Upload'' and follow the instructions to transfer your audio to the HPC environment. Please note that the example here contains a single file, but that Whisper can transcribe multiple files in the same job. Feel free to upload as many audio files as needed. The only limitation you have is that the maximum runtime of Whisper, as it is set up now, covers about 20 hours of interviews. If you need to transcribe more than this amount of time, please consider splitting the data into two separate batches and launching two separate jobs. | You only need to do this once. |
| |
| {{ :dcc:itsol:whisper:hb_portal_4a.png?direct&900 | }} | - In the web portal, open the ''%%Files%%'' menu and choose your ''%%scratch%%'' directory, which is ''%%/scratch/%%'' followed by your username, for example ''%%/scratch/p123456%%''. The file browser opens. |
| | - Click ''%%New Directory%%'', type ''%%whisper%%'', and click ''%%OK%%''. |
| | - Click the new ''%%whisper%%'' directory to open it. |
| | - Create an ''%%input%%'' and an ''%%output%%'' directory inside it in the same way. |
| | |
| | Your ''%%scratch%%'' directory now contains: |
| | |
| | <code> |
| | whisper/ |
| | ├── input/ |
| | └── output/ |
| | </code> |
| | |
| | Use these lowercase names. The form then proposes the right directories for you, inside the ''%%scratch%%'' directory you opened in step 1. If your recordings live elsewhere, for example in a group ''%%scratch%%'' directory, check the ''%%Folder for input files%%'' and ''%%Folder for output files%%'' fields when you fill in the form; see [[dcc:itsol:whisper:running#input_and_output_directories|Input and Output Directories]]. |
| | |
| | Put only the recordings for the current run in ''%%whisper/input%%''. **The app hands every entry in the input directory to Whisper, so any other file makes the job fail**. The job script is the usual cause for this: leave ''%%Script Location%%'' pointing at ''%%whisper/%%'', never at ''%%whisper/input%%'', because the script and the job log would then be treated as recordings. |
| | |
| | {{ :dcc:itsol:whisper:create_dirs.webm?800x450 |}} |
| | |
| | ===== Upload recordings ===== |
| | |
| | - In the file browser, open ''%%whisper%%'' and then ''%%input%%''. |
| | - Click ''%%Upload%%''. |
| | - Drag the audio or video files into the window, or click ''%%browse files%%'' and select them. |
| | - Click the green ''%%Upload%%'' button, which shows the number of files, for example ''%%Upload 3 files%%''. |
| | - Wait until every upload has finished before you start the job. |
| | |
| | {{ :dcc:itsol:whisper:upload.webm?800x450 |}} |
| | |
| | <WRAP half column>[[dcc:itsol:whisper|← Previous: Whisper Guide]]</WRAP> |
| | <WRAP half column rightalign>[[dcc:itsol:whisper:running|Next: Run a Whisper Job →]]</WRAP> |
| | <WRAP clear></WRAP> |
| |
| [[dcc:itsol:whisper:scripts| → Move to the next step]] | |