mirdata icon indicating copy to clipboard operation
mirdata copied to clipboard

Issues with tag_to_jams

Open bmcfee opened this issue 2 years ago • 2 comments

I noticed a couple of small bugs with tag_to_jams (permalink):

https://github.com/mir-dataset-loaders/mirdata/blob/e591c5411c41591e8606812df869dca1ad52ee0f/mirdata/jams_utils.py#L556-L576

  1. The documentation copypastad lyrics
  2. There is no way to set the confidence value for an observation
  3. There is no way to set the time or durations

1 is easy enough to fix. 2 and 3 seem like the same problem, caused by a lack of an underlying tag datastructure (only string is supported).

bmcfee avatar Nov 09 '21 01:11 bmcfee

Thanks for opening this issue Brian, I'll be working on this and refactoring of jams' converters in #546. One quick check: so far we were treating tags as immutable str for the whole track, so I don't get why we need time if it's always going to be time=0.0 (beginning of track). What am I missing?

magdalenafuentes avatar Nov 19 '21 00:11 magdalenafuentes

Tags can apply to any time interval, so this should be exposed here. It's not a huge deal for most data sets where tags are static, but I see no reason not to support it.

Openmic really does need access to the confidence field though.

bmcfee avatar Nov 19 '21 02:11 bmcfee