[MRG] Add MultipitchData (#182)

* fixes #101 * run black
mir-dataset-loaders · Mar 6, 2020 · 1baecfb · 1baecfb
1 parent ab629f3
commit 1baecfb
Show file tree

Hide file tree

Showing 3 changed files with 34 additions and 42 deletions.
diff --git a/mirdata/medleydb_melody.py b/mirdata/medleydb_melody.py
@@ -120,7 +120,7 @@ def __repr__(self):
             + "is_instrumental={}, n_sources={}, "
             + "melody1=F0Data('times', 'frequencies', confidence'), "
             + "melody2=F0Data('times', 'frequencies', confidence'), "
-            + "melody3=F0Data('times', 'frequencies', confidence'))"
+            + "melody3=MultipitchData('times', 'frequencies', confidence'))"
         )
         return repr_string.format(
             self.track_id,
@@ -156,12 +156,9 @@ def audio(self):
         return librosa.load(self.audio_path, sr=None, mono=True)
 
     def to_jams(self):
+        # jams does not support multipitch, so we skip melody3
         return jams_utils.jams_converter(
-            f0_data=[
-                (self.melody1, 'melody1'),
-                (self.melody2, 'melody2'),
-                (self.melody3, 'melody3'),
-            ],
+            f0_data=[(self.melody1, 'melody1'), (self.melody2, 'melody2')],
             metadata=self._track_metadata,
         )
 
@@ -268,17 +265,17 @@ def _load_melody3(melody_path):
     if not os.path.exists(melody_path):
         return None
     times = []
-    freqs = []
+    freqs_list = []
+    conf_list = []
     with open(melody_path, 'r') as fhandle:
         reader = csv.reader(fhandle, delimiter=',')
         for line in reader:
             times.append(float(line[0]))
-            freqs.append([float(v) for v in line[1:]])
+            freqs_list.append([float(v) for v in line[1:]])
+            conf_list.append([float(float(v) > 0) for v in line[1:]])
 
     times = np.array(times)
-    freqs = np.array(freqs)
-    confidence = (freqs > 0).astype(float)
-    melody_data = utils.F0Data(times, freqs, confidence)
+    melody_data = utils.MultipitchData(times, freqs_list, conf_list)
     return melody_data
 
 

diff --git a/mirdata/utils.py b/mirdata/utils.py
@@ -154,6 +154,10 @@ def validator(dataset_index, data_home, silence=False):
 
 F0Data = namedtuple('F0Data', ['times', 'frequencies', 'confidence'])
 
+MultipitchData = namedtuple(
+    'MultipitchData', ['times', 'frequency_list', 'confidence_list']
+)
+
 LyricData = namedtuple(
     'LyricData', ['start_times', 'end_times', 'lyrics', 'pronunciations']
 )

diff --git a/tests/test_medleydb_melody.py b/tests/test_medleydb_melody.py
@@ -61,7 +61,7 @@ def test_track():
 
     assert type(track.melody1) is utils.F0Data
     assert type(track.melody2) is utils.F0Data
-    assert type(track.melody3) is utils.F0Data
+    assert type(track.melody3) is utils.MultipitchData
 
     y, sr = track.audio
     assert sr == 44100
@@ -74,7 +74,7 @@ def test_track():
         + " genre=Classical, is_excerpt=True, is_instrumental=True, "
         + "n_sources=18, melody1=F0Data('times', 'frequencies', confidence'),"
         + " melody2=F0Data('times', 'frequencies', confidence'), "
-        + "melody3=F0Data('times', 'frequencies', confidence'))"
+        + "melody3=MultipitchData('times', 'frequencies', confidence'))"
     )
     assert track.__repr__() == repr_string
 
@@ -147,42 +147,33 @@ def test_load_melody3():
     melody_data = medleydb_melody._load_melody3(melody_path)
 
     # check types
-    assert type(melody_data) == utils.F0Data
+    assert type(melody_data) == utils.MultipitchData
     assert type(melody_data.times) is np.ndarray
-    assert type(melody_data.frequencies) is np.ndarray
-    assert type(melody_data.confidence) is np.ndarray
+    assert type(melody_data.frequency_list) is list
+    assert type(melody_data.confidence_list) is list
 
     # check values
     assert np.array_equal(
         melody_data.times,
         np.array([0.046439909297052155, 0.052244897959183675, 0.1219047619047619]),
     )
-    assert np.array_equal(
-        melody_data.frequencies,
-        np.array(
-            [
-                [0.0, 0.0, 497.01600000000002, 0.0, 0.0],
-                [965.99199999999996, 996.46799999999996, 497.10599999999999, 0.0, 0.0],
-                [
-                    987.32000000000005,
-                    987.93200000000002,
-                    495.46800000000002,
-                    495.29899999999998,
-                    242.98699999999999,
-                ],
-            ]
-        ),
-    )
-    assert np.array_equal(
-        melody_data.confidence,
-        np.array(
-            [
-                [0.0, 0.0, 1.0, 0.0, 0.0],
-                [1.0, 1.0, 1.0, 0.0, 0.0],
-                [1.0, 1.0, 1.0, 1.0, 1.0],
-            ]
-        ),
-    )
+    assert melody_data.frequency_list == [
+        [0.0, 0.0, 497.01600000000002, 0.0, 0.0],
+        [965.99199999999996, 996.46799999999996, 497.10599999999999, 0.0, 0.0],
+        [
+            987.32000000000005,
+            987.93200000000002,
+            495.46800000000002,
+            495.29899999999998,
+            242.98699999999999,
+        ],
+    ]
+
+    assert melody_data.confidence_list == [
+        [0.0, 0.0, 1.0, 0.0, 0.0],
+        [1.0, 1.0, 1.0, 0.0, 0.0],
+        [1.0, 1.0, 1.0, 1.0, 1.0],
+    ]
 
     # load a file which doesn't exist
     melody_data_none = medleydb_melody._load_melody3('fake/file/path')