African American Vernacular Media Audio dataset

Home » Case Study » African American Vernacular Media Audio dataset

Project Overview:

Objective

The “African American Vernacular Media Audio Dataset” project was designed to curate a comprehensive dataset for training machine learning models to recognize and analyze African American Vernacular English (AAVE) in media content. This dataset serves as a valuable resource for linguistic research, cultural studies, and natural language processing applications.

Scope

The project involved the collection of audio clips featuring instances of African American Vernacular English (AAVE) speech from various media sources, including movies, TV shows, interviews, and online videos. After collecting the data, we meticulously annotated it to identify AAVE instances, thus providing valuable insights into the use of AAVE in media.

Sources

Movies and TV Shows: Scenes featuring characters speaking in AAVE.
Interviews: Conversations and interviews with African American individuals where AAVE is spoken.
Online Videos: Social media clips, YouTube videos, and online content showcasing AAVE usage.

Data Collection Metrics

Total Audio Clips Collected: 20,000 clips
Media Sources: 15,000 clips
Interviews: 3,000 clips
Online Videos: 2,000 clips

Annotation Process

Stages

Linguistic Annotation: Trained linguists have meticulously identified and marked segments where AAVE is spoken within each audio clip. This detailed annotation ensures accurate documentation and analysis.
Metadata Logging: Furthermore, we have recorded metadata, including the source of the clip, context, and relevant cultural information. This comprehensive approach enriches the dataset.

Annotation Metrics

Audio Clips with AAVE Labels: 15,000 clips
Metadata Annotations: 10,000c

Quality Assurance

Stages

Annotation Verification: Linguists and cultural experts reviewed and verified the accuracy of AAVE annotations. Additionally, they ensured that every detail was meticulously checked for authenticity.
Data Quality Control: Rigorous checks were conducted to remove low-quality or noisy clips, ensuring a clean and reliable dataset. Consequently, we achieved a high standard of data integrity.
Data Security: We prioritized the protection of sensitive linguistic data. Moreover, we adhered to privacy regulations and obtained necessary permissions when required, ensuring comprehensive data security.

QA Metrics

Annotation Validation Cases: 2,000
Data Cleansing: Removal of low-quality or irrelevant clips

Conclusion

The “African American Vernacular Media Audio Dataset” empowers linguists, researchers, and developers to study African American Vernacular English in media. It provides a substantial collection of annotated audio clips, opening avenues for understanding AAVE’s cultural and linguistic nuances in various media contexts. Consequently, it enables the development of machine learning models, cultural studies, and linguistic research. This contribution fosters a deeper appreciation of language diversity and cultural representation in media content.