gilkeyio/inclusive_voice

Dataset Card for Inclusive Voice

A dataset curated from Mozilla Common Voice containing equal representation from male, female, and other

Dataset Details

Dataset Sources

Mozilla Common Voice

• Homepage: https://commonvoice.mozilla.org/en/datasets
• Repository: https://github.com/common-voice/common-voice
• Paper: https://arxiv.org/abs/1912.06670

Uses

Direct Use

This dataset is intended for use in training and evaluating voice recognition models, particularly for applications in gender classification.

Out-of-Scope Use

The dataset is not suitable for applications requiring a wide range of dialects or non-English languages. It is also not designed for identifying fine-grained demographic characteristics beyond gender. You agree not to attempt to identify the individuals in the dataset.

Dataset Structure

Data Fields

• client_id: A unique identifier for each participant.
• audio: Audio recording of a spoken sentence.
• sentence: The text content of the audio recording.
• age: Categorical age group of the participant.
• gender: Gender of the participant (balanced representation).
• accent: Accent information of the participant, if available.

Splits

• Train: 6,075 examples (2,025 of each gender).
• Test: 1,812 examples (604 of each gender).

Dataset Creation

Curation Rationale

The dataset was specifically curated to address gender imbalance in voice recognition datasets, aiming to provide a balanced representation of male, female, and non-binary voices.

Source Data

Data Collection and Processing

Derived from Mozilla's Common Voice project, the dataset underwent additional processing to balance gender representation and limit the dominance of any single voice.

Bias, Risks, and Limitations

Despite efforts to balance gender representation, the dataset may still have biases, such as over-representation of certain accents or age groups. It's not comprehensive in representing all possible variations in English speech. Since the original dataset contains relatively few female and even fewer non-binary voices, the dataset becomes quite small when filtered down.

Citation

The original Common Voice citation:

@inproceedings{commonvoice:2020,
  author = {Ardila, R. and Branson, M. and Davis, K. and Henretty, M. and Kohler, M. and Meyer, J. and Morais, R. and Saunders, L. and Tyers, F. M. and Weber, G.},
  title = {Common Voice: A Massively-Multilingual Speech Corpus},
  booktitle = {Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020)},
  pages = {4211--4215},
  year = 2020
}

Dataset Card Authors

Kim Gilkey

Dataset Card Contact

kim@gilkey.io