Skip to content

Dataset · upb-nlp

300 Hours of Romanian Speech by 300 Speakers

upb-nlp/echo

Crowdsourced Romanian audio corpus for training speech recognition systems.

What's inside

The dataset includes over 300 hours of audio recordings with text transcripts in Romanian (from 100,000 to 1 million segments) spoken by 300 different speakers. The data was collected via the Echo crowdsourcing platform. The corpus is open strictly for non-commercial research purposes.

What you can do with it

  • Training and fine-tuning automatic speech recognition (ASR) models for the Romanian language
  • Adapting multilingual models like Whisper to Romanian speech
Tasks
Automatic speech recognition
Languages
Romanian
Size
100K–1M rows
Downloads this month
250
Likes
5
Access
on request on Hugging Face

How to load it

from datasets import load_dataset

ds = load_dataset("upb-nlp/echo")