ADIMA / README.md
callmesan's picture
Update README.md
46ec391 verified
metadata
dataset_info:
  features:
    - name: path_to_audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: abuse
      dtype: string
    - name: language
      dtype: string
  splits:
    - name: train
      num_bytes: 5192187842.424
      num_examples: 8128
    - name: test
      num_bytes: 2347579907.564
      num_examples: 3647
  download_size: 5849025117
  dataset_size: 7539767749.988
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
task_categories:
  - audio-classification
language:
  - ta
  - bn
  - gu
  - hi
  - or
  - ml
  - kn
  - pa
tags:
  - audio-abuse
size_categories:
  - 10K<n<100K

ADIMA is a dataset by ShareChat Inc. Dataset Statistcs and other information in the paper. I am in no way affiliated with ShareChat. Just helping other users in Open Science.

Cite them with the following:

@INPROCEEDINGS{9746718,
  author={Gupta, Vikram and Sharon, Rini and Sawhney, Ramit and Mukherjee, Debdoot},
  booktitle={ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
  title={ADIMA: Abuse Detection In Multilingual Audio}, 
  year={2022},
  volume={},
  number={},
  pages={6172-6176},
  abstract={Abusive content detection in spoken text can be addressed by performing Automatic Speech Recognition (ASR) and leveraging advancements in natural language processing. However, ASR models introduce latency and often perform sub-optimally for abusive words as they are underrepresented in training corpora and not spoken clearly or completely. Exploration of this problem entirely in the audio domain has largely been limited by the lack of audio datasets. Building on these challenges, we propose ADIMA, a novel, linguistically diverse, ethically sourced, expert annotated and well- balanced multilingual abuse detection audio dataset comprising of 11,775 audio samples in 10 Indic languages spanning 65 hours and spoken by 6,446 unique users. Through quantitative experiments across monolingual and cross-lingual zeroshot settings, we take the first step in democratizing audio based content moderation in Indic languages and set forth our dataset to pave future work. Dataset and code are available at: https://github.com/ShareChatAI/Adima},
  keywords={Training;Ethics;Codes;Speech coding;Conferences;Buildings;Signal processing;Abusive Content Detection;Multilingual Audio Analysis;Indic Dataset;Crosslingual Audio Analysis},
  doi={10.1109/ICASSP43922.2022.9746718},
  ISSN={2379-190X},
  month={May},}