Some datasets are indeed like this, see eg Common Voice - https://voice.mozilla.org/en

In general these things are open source, so you can always contribute an improved version of the dataset. But as another commenter said having relatively static ones is also important for benchmarking purposes.