metadata
dataset_info:
features:
- name: sample_id
dtype: int64
- name: noise_type
dtype: string
- name: clean_audio
dtype:
audio:
sampling_rate: 16000
- name: baseline_predicted_audio
dtype:
audio:
sampling_rate: 16000
- name: grpo_predicted_audio
dtype:
audio:
sampling_rate: 16000
- name: clean_transcription
dtype: string
- name: baseline_transcription
dtype: string
- name: grpo_transcription
dtype: string
- name: wer_baseline
dtype: float64
- name: wer_grpo
dtype: float64
- name: wer_diff
dtype: float64
splits:
- name: train
num_bytes: 5050134
num_examples: 10
download_size: 4601636
dataset_size: 5050134
configs:
- config_name: default
data_files:
- split: train
path: data/train-*