Know what goes in.
A model’s results begin with its data. This is where we’ll document sources, preparation, and evaluation.
Preparing our first releaseDocumentation
Documentation
before downloads.
No datasets are published here yet. Release entries will distinguish data we can redistribute from third-party sources that must be obtained separately.
Follow the Rivet projectWhat we intend to document
The details matter.
01
Provenance
Source, version, language, collection context, and the rights associated with each source.
02
Preparation
Cleaning, segmentation, filtering, and the steps that turn source audio into training examples.
03
Evaluation
How the splits are defined, what is held out, and how overlap is checked.
04
Limitations
Coverage gaps, known biases, exclusions, and the uses an evaluation does not support.
Looking for reuse terms?
Read about licensing