Neural Audio Fingerprint Dataset For High-Specificity Retrieval

Encompassing music sources, background noise, and impulse-response samples used for training a neural audio fingerprinting model described in a 2021 arXiv paper. The audio files are 16-bit PCM mono WAV format with an 8000 Hz sampling rate. The dataset is hosted on Hugging Face and was last updated in April 2022.

Use Cases

• Train a contrastive learning model for high-specificity audio retrieval using the provided music and noise samples.
• Benchmark audio fingerprinting algorithms against the impulse-response samples to test robustness.
• Use the 8000 Hz mono WAV files to develop or evaluate neural network architectures for audio representation learning.

Strengths

• Includes multiple audio categories: music sources, background noise, and impulse-response samples.
• Audio files are provided in a consistent 16-bit PCM mono WAV format at 8000 Hz.
• Dataset is associated with a peer-reviewed research paper (arXiv:2010.11910) detailing its use.

Limitations

• The exact number of audio files, total size, and specific columnar metadata are unknown.
• The dataset's temporal and geographic coverage for the music sources is not specified.
• License information is not provided, which may restrict usage.

Provenance

Source

Created by Sungkyun Chang (arch-raven) and hosted on Hugging Face.

Collection Method

Compiled for the research paper 'Neural Audio Fingerprint for High-specific Audio Retrieval based on Contrastive Learning'.

Freshness

Last updated on 2022-04-05.