Audiocaption: Audio-Text Pairs and Baselines for Audio Description

This audio-text dataset provides paired audio signals and descriptive captions for the first Audiocaption task, released by RicherMans in 2024. It serves as a benchmark for automated audio description systems and includes baseline code for performance evaluation.

Use Cases

• Training neural networks to map audio signals to natural language captions
• Evaluating the accuracy of the provided baseline models on the Audiocaption task
• Analyzing the relationship between acoustic features and descriptive text labels

Strengths

• Includes baseline models for the Audiocaption task
• Released under the MIT license

Limitations

• The total record count and dataset size are not documented in the source metadata
• Geographic and environmental diversity of the audio recordings is unknown

Provenance

Source

RicherMans GitHub repository

Freshness

Last updated July 2024.