A Pilot Study on the JASMIN-CGN Corpus Using the Whisper Model

Description

We test and study the variation in speech recognition of fine-tuned versions of Whisper on children, elderly and non-native Dutch speech from the JASMIN-CGN corpus.

Requirements

Make sure you have JASMIN-CGN downloaded (download from here: [https://taalmaterialen.ivdnt.org/download/tstc-jasmin-spraakcorpus/])

How to use the files:

• First you will remove the silences of the data
• Then you need to make the data into 30 second chunks
• You need to make a csv from the ort files! and convert to UTF-8
• divide your datasets
• NOW WE CAN TRAIN
• AND TEST!