Raw Audio Waveforms for Single-Instrument Piano Music

YouTubeMix is a raw audio waveform dataset used for training music generation models. It contains single-instrument piano music derived from a specific YouTube video audio track. The dataset was created by krandiash and was last updated in February 2022.

Use Cases

• Train state-space models on raw audio waveforms for music generation, as referenced in the paper 'It's Raw! Audio Generation with State-Space Models'.
• Use as a source of single-instrument piano music for small-scale model training, as described in the dataset's purpose.
• Benchmark audio generation models against the dataset used in the original SampleRNN GitHub repository from the Deep Sound Project.

Strengths

• Dataset is associated with a specific research paper ('It's Raw! Audio Generation with State-Space Models'), providing a clear academic context.
• The audio source is a single, identified YouTube video, ensuring consistency in the raw audio material.
• Dataset is tagged with an Arxiv identifier (2202.09729), linking it to formal research.

Limitations

• Dataset size is categorized as 'n1 K', indicating it contains only a few thousand audio samples, which is small for deep learning.
• The dataset consists of audio from a single YouTube video, limiting musical diversity and introducing potential source bias.
• The dataset was last updated in February 2022, which may be considered stale for rapidly evolving audio generation research.

Provenance

Source

Audio track from a specific YouTube video (https://www.youtube.com/watch?v=EhO_MrRfftU).

Collection Method

Derived from the audio of a single YouTube video, originally used in the SampleRNN GitHub repository.

Freshness

Last updated on 2022-02-22.

Geography

Tagged with region 'us', but specific geographic coverage of the audio content is unknown.