metadata
dataset_info:
features:
- name: ID
dtype: string
- name: audio
dtype:
audio:
sampling_rate: 16000
- name: country
dtype: string
splits:
- name: train
num_bytes: 1783821218.4609375
num_examples: 12900
- name: validation
num_bytes: 1746232603.9765625
num_examples: 12700
download_size: 3533048242
dataset_size: 3530053822.4375
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
To participate in the NADI 2025 Spoken Dialect ID challenge please make sure you have visited the main NADI 2025 page link, and sign the participation form on CodaBench. Ensure your email + contact information matches your Huggingface Access request email.
This is the `adaptation' split for the NADI 2025 Spoken Dialect ID task. As an adaptation split, the idea is to use an existing external dataset (e.g. ADI-17) for the main training, and then use this split for fine-tuning ('train') and validation.
This is a version of the nadi-asr dataset, without overlapping speakers between train / validation, and reformatted for ease of use in training for dialect ID.