Companion code to mirdata issue #236

bug-236.py

from mirdata import orchset
import os

def print_duration(mirdata_track):
    """This function measures the duration of a mirdata
    Track by converting to JAMS and reading the file_metadata.
    It runs only on mirdata master and returns an AssertionError
    in mirdata v0.2-beta"""
    jam = mirdata_track.to_jams()
    assert jam.validate()
    print(jam.file_metadata.duration)
    

def data_augmentation(orchset_home, track_id, flag=True):
    """Here's a borrowed piece of code that I don't fully understand,
    but I need it for my research. What I can tell is that it takes
    two strings as arguments. This has nothing to do with mirdata
    and only runs shell commands, so what could go wrong?"""
    # Narrator: do NOT run this code unless you know what you're doing
    # This function destroys the ORCHSET dataset by trimming the beginning
    # of a wave file by 100 milliseconds.
    # Because this function doesn't adjust annotations accordingly,
    # evaluating a melody estimator on this modified ORCHSET would be
    # a catastrophe.
    audio_path = os.path.join(
        orchset_home, "audio", "mono", track_id + ".wav")
    if flag:
        temp_path = os.path.join(orchset_home, "temp.wav")
        os.system(" ".join(["sox", audio_path, temp_path, "trim", "0.1"]))
    os.remove(audio_path)
    if flag:
        os.rename(temp_path, audio_path)
    
### Now, suppose I just installed mirdata and ran the README code.
# I'm going to run validate, just to be sure; but i'll run it only
# once because it takes a long time.
#orchset.validate()
# mirdata v0.2-beta -> ({}, {})
# mirdata master -> ({}, {})
# Nice, all checksums are correct!

# Let's make an ORCHSET Track!
track_id = "Beethoven-S3-I-ex2"
track = orchset.Track(track_id)

# What is its duration?
print_duration(track)
# -> 10.282879818594104
# OK, noted.

# Now, let me experiment with data augmentation ...
# I won't be using mirdata for this.
vincent_orchset_home = "/Users/vl238/mir_datasets/Orchset/"
data_augmentation(vincent_orchset_home, track_id)


# ... a few moments later ...
# What is the duration of my Track again?
print_duration(track)
# -> 10.182879818594104


# Wait, what?
print_duration(track)
# -> 10.182879818594104

# What happened? What did I do? I haven't been using mirdata at all
# since I last printed duration! Is it a bug in print_duration??
print_duration(track)
# -> 10.182879818594104

# Doesn't seem like it. Weird.


# ... a few moments later ...
# Perhaps I should try running data_augmentation with flag=False?
data_augmentation(vincent_orchset_home, track_id, flag=False)

# Let me try to read the duration again
print_duration(track)
# -> ERROR. Excerpt of the backtrace below
# 
#----> 9     jam = mirdata_track.to_jams()
#     10     assert jam.validate()
#     11     print(jam.file_metadata.duration)
#
#~/mirdata/mirdata/orchset.py in to_jams(self)
#    196         metadata = {k: v for k, v in self._track_metadata.items() if v is not None}
#    197         metadata['duration'] = librosa.get_duration(
#--> 198             self.audio_mono[0], self.audio_mono[1]
#    199         )
#    200         return jams_utils.jams_converter(
#
#~/mirdata/mirdata/orchset.py in audio_mono(self)
#    185     def audio_mono(self):
#    186         """(np.ndarray, float): mono audio signal, sample rate"""
#--> 187         return load_audio_mono(self.audio_path_mono)
#    188 
#    189     @property

#~/mirdata/mirdata/orchset.py in load_audio_mono(audio_path)
#    214 
#    215     """
#--> 216     return librosa.load(audio_path, sr=None, mono=True)
#    217 
#    218 

# Argh! Where is the bug??
# I don't think the bug is in print_duration. I'm just calling the to_jams()
# method and printing the result.
# Also, if I run the same script on guitarset, print_duration is constant.
# Is it something with the ORCHSET loader?