A small hobby project trained in a Kaggle notebook using their free P100 GPUs. Was curious about if you could train whisper-tiny to perform decently if you specialized it for a single language, i.e. danish in this case. The TL;DR is that the results are not great :)

from transformers import pipeline

pipe = pipeline("automatic-speech-recognition", 
                model="rasgaard/whisper-tiny.da")

Downloads last month: 9

Safetensors

Model size

57.7M params

Tensor type

F32

Model tree for rasgaard/whisper-tiny.da

Base model

openai/whisper-tiny

Finetuned

(1829)

this model

Dataset used to train rasgaard/whisper-tiny.da

Article mentioning rasgaard/whisper-tiny.da

Free Fine-tuning of Whisper-tiny for Danish