Aac Datasets: PyTorch-Compatible Audio Captioning Resources

Labbeti developed this collection of audio captioning datasets specifically for integration with PyTorch deep learning frameworks, with the most recent update in July 2025. The data facilitates the development of models that generate natural language descriptions from audio inputs. It serves as a centralized resource for researchers working on cross-modal audio-to-text tasks.

Use Cases

• Training sequence-to-sequence models to generate text captions from raw audio signals
• Evaluating audio encoder architectures within PyTorch-based pipelines
• Benchmarking automated audio description systems using the provided caption labels

Strengths

• MIT licensed for permissive commercial and academic use
• Pre-configured for PyTorch deep learning environments
• Specialized focus on the niche audio-to-text captioning domain

Limitations

• Unknown record count and dataset scale
• Lack of detailed metadata regarding audio sampling rates or bit depth
• Source of the audio recordings is not explicitly documented

Provenance

Source

Labbeti (GitHub)

Freshness

Updated as of July 2025.