Add spanish carrion crows dataset#255
Conversation
|
|
||
| audio, sr = read_audio(audio_path) | ||
| # Should all be mono | ||
| # audio = audio_stereo_to_mono(audio, mono_method="average").astype(np.float32) |
There was a problem hiding this comment.
you can just remove the line if its already mono (maybe test with an assert in your tests)
There was a problem hiding this comment.
Ok removed and added a test. I realized that I took out the .astype(np.float32) as well when I did this. Is it important to keep float32 for any reason? (The tests check for float64 so it's consistent with itself right now)
There was a problem hiding this comment.
well float32 saves memory (especially GPU).. and float64 doesn't really offer much more useful precision for audio ?
the other thing is that every dataset (I believe) in esp_data is enforcing float32 so having float64 in just this one will break experiments where this dataset is concatenated / chained with others in a training loop ... wdyt ?
There was a problem hiding this comment.
Okay, I'll leave in the conversion to float32
GaganNarula
left a comment
GaganNarula
left a comment
There was a problem hiding this comment.
some comments to discuss but in principle it looks good!
|
I think I addressed all the comments so far now. Jules said it would be useful to add call types, I will do that in a second version |
2 participants
I'm adding the unsynchronized, audio-only version of the carrion crow biologger dataset. It is a detection dataset (Voxaboxen bounding boxes). I followed WABAD/ArcticBirdSounds as templates.