Skip to content

Dataset · s3prl

Speech model evaluation benchmark (SUPERB)

s3prl/superb

English audio and text data for evaluating universal speech processing models.

What's inside

The dataset includes 100k to 1M examples in English audio and text modalities. Created for the SUPERB leaderboard, testing a single base model's ability to solve various speech tasks with minimal fine-tuning. Specifically, data includes phone recognition (PR) and other ASR and sound classification tasks.

What you can do with it

  • Comparing pretrained audio encoders (like Wav2Vec2 or HuBERT) on a unified benchmark
  • Evaluating automatic speech recognition (ASR) and phone systems
  • Testing sound classification models
Tasks
Automatic speech recognition, Audio classification
Languages
English
Size
100K–1M rows
Downloads this month
1.3K
Likes
34

How to load it

from datasets import load_dataset

ds = load_dataset("s3prl/superb")