Add fma keys dataset #625
Add fma keys dataset #625
Conversation
There was a problem hiding this comment.
Hello @stellaywong, sorry for the late late late reply... we were very busy with soundata and its corresponding JOSS submission. Thank you for your contribution to mirdata :) We are targeting a release for next month, and we would like this loader to be included, glad to help on that! The PR is really good, it only needs a bit of refinement, the module docstring, and the documentation (find here the instructions for that!). Let us know if you need any help!
| # librosa has problems reading FMA mp3s without clamping down to the second. | ||
| duration = librosa.get_duration(path=fhandle) | ||
| return librosa.load(fhandle, sr=None, mono=True, duration=floor(duration)) |
There was a problem hiding this comment.
wow is that so? Can we track this error and see why it happens? I'll take a look. Nice workaround for now :)
There was a problem hiding this comment.
Yes, I'll dig more into it. We didn't see this with other mp3s, just with ones coming from the FMA dataset.
There was a problem hiding this comment.
Thanks! Let me know if yu want us to also take a look at that issue :)
|
@stellaywong thanks for contributing to |
This adds support for the FMA Keys dataset, which is a new dataset for the evaluation of key detection containing 340 hours (5489 songs) of song-level key and mode annotations, spread across 17 genres.
|
Thank you for the reviews! I've made the changes and pushed the updates. It looks like the PR build is stuck waiting for status to be reported? Do you know how we can kick start the build? |
|
Hey @stellaywong, thanks for the update :) Sorry that I was afk for a few days. OK, that looks quite good, you are missing a few things that I am gonna list in my next review, but the loader looks good! Hope that you will have some time to do the final updates... thanks for the patience! |
There was a problem hiding this comment.
Alright, getting way closer! Thanks again for your patience and collaboration :) I just left few indications of the things your PR is missing. Please let me know if you have any questions. Thanks @stellaywong!
| return jams_utils.jams_converter( | ||
| metadata=self._track_metadata, | ||
| ) |
There was a problem hiding this comment.
You jams_converter is missing the audio_path, so it cannot compute needed attributes of the corresponding audio. See this exemple:
return jams_utils.jams_converter(
audio_path=self.audio_path,
metadata={
"instrument": self.instrument,
"genre": self.genre,
"drum": self.drum,
"train": self.train,
},
)
That will be needed for the tests to pass!
| """ | ||
| FMA Keys Dataset Loader | ||
|
|
||
| .. admonition:: Dataset Info | ||
| :class: dropdown |
There was a problem hiding this comment.
alright, now the docstrings looks quite good. However, you need to update some files in the documentation source code to make this visible to users.
-
docs/source/mirdata.rst: here, you need to add your dataset as it is done for the rest! -
docs/source/table.rst: here, you need to add your dataset as it is done for the rest as well. Add the number of tracks, available annotations, data availability, licensing, etc. -
docs/source/quick_reference.rst: if your dataset has some annotation type of information that is not available in quick reference to link, please add it there, and therefore, users will be able to click the annotations in the table and read more info!
| spotify_uri (str): Spotify URI if available | ||
| key (str): path to the track's audio file | ||
| mode (str): path to the track's audio file | ||
| key_number (int): path to the track's audio file | ||
| mode_number (int): path to the track's audio file | ||
| audio (str): path to the track's audio file |
There was a problem hiding this comment.
most of them seem to have path to the track's audio file as description. Please update! also please change audio for audio_path` .
| # librosa has problems reading FMA mp3s without clamping down to the second. | ||
| duration = librosa.get_duration(path=fhandle) | ||
| return librosa.load(fhandle, sr=None, mono=True, duration=floor(duration)) |
There was a problem hiding this comment.
Thanks! Let me know if yu want us to also take a look at that issue :)
This adds support for the FMA Keys dataset, which is a new dataset for the evaluation of key detection containing 340 hours (5489 songs) of song-level key and mode annotations, spread across 17 genres.