Datasets:
Tasks:
Automatic Speech Recognition
Multilinguality:
multilingual
Language Creators:
crowdsourced
Annotations Creators:
crowdsourced
Source Datasets:
extended|common_voice
ArXiv:
Tags:
License:
load_dataset() function whit split='test' is downloading all data.
#5
by
SantiagoMoreno-UdeA
- opened
When you use the load_dataset("mozilla-foundation/common_voice_13_0", "es", split="test") the function download and generates all splits in the data, not only the test set.
The best option here would be to stream the dataset and load samples on the fly:
from datasets import load_dataset
from tqdm import tqdm
dataset = load_dataset("mozilla-foundation/common_voice_13_0", "es", split="test", streaming=True)
# stream individual samples on the fly
for sample in tqdm(dataset):
audio = sample["audio"]
text = sample["sentence"]
# do whatever you need to do...
You can also use a streaming dataset with a torch data loader for batched inference (see https://huggingface.co/datasets/mozilla-foundation/common_voice_13_0#streaming)
Thanks Sanchit!