---
dataset_info:
  features:
  - name: audio
    dtype: audio
  - name: transcription
    dtype: string
  - name: duration
    dtype: float64
  - name: sr
    dtype: int64
  - name: wav2vec2pred
    dtype: string
  - name: wer
    dtype: float64
  - name: wps
    dtype: float64
  - name: is_better
    dtype: bool
  splits:
  - name: commonvoice
    num_bytes: 26613419533.408
    num_examples: 963636
  download_size: 58948504311
  dataset_size: 61107679180.05001
configs:
- config_name: default
  data_files:
  - split: commonvoice
    path: data/commonvoice-*
task_categories:
- automatic-speech-recognition
language:
- bn
size_categories:
- 1M<n<10M
---

# Open Large Bengali ASR Data

This is a collection of publicly available ASR data for Bengali. It contains **5000 hours** of audio. We have a filtering column called *is_better* to filter good-quality audio from the corpus. It is set based on the *wer* between original transcription and prediction taken from a Bengali-Wav2Vec2 model and *word-per-second (wps)*. 

# Datasets:

- [commonvoice](https://www.kaggle.com/competitions/bengaliai-speech/data)