fixing stereo handling in sisdr

benjohn18 · Nov 22, 2019 · dde303e · dde303e
1 parent b1fc5eb
commit dde303e
Show file tree

Hide file tree

Showing 3 changed files with 4 additions and 3 deletions.
diff --git a/.gitignore b/.gitignore
@@ -1,5 +1,6 @@
 #### joe made this: http://goel.io/joe
 
+brouillon.py
 *.png
 *.wav
 *.jpg

diff --git a/speechmetrics/absolute/mosnet/model.py b/speechmetrics/absolute/mosnet/model.py
@@ -89,7 +89,7 @@ def __init__(self, window, hop=None):
 
     def test_window(self, audios, rate):
         # stft. D: (1+n_fft//2, T)
-        linear = librosa.stft(y=audios[0],
+        linear = librosa.stft(y=np.asfortranarray(audios[0]),
                               n_fft=self.FFT_SIZE,
                               hop_length=self.HOP_LENGTH,
                               win_length=self.WIN_LENGTH,

diff --git a/speechmetrics/relative/sisdr.py b/speechmetrics/relative/sisdr.py
@@ -13,8 +13,8 @@ def test_window(self, audios, rate):
         # and one estimate.
         # see original code here: https://github.com/sigsep/bsseval/issues/3#issuecomment-494995846
         eps = np.finfo(audios[0].dtype).eps
-        reference = audios[1].reshape(audios[1].shape[0], -1)
-        estimate = audios[0].reshape(audios[0].shape[0], -1)
+        reference = audios[1].reshape(audios[1].size, 1)
+        estimate = audios[0].reshape(audios[0].size, 1)
         Rss = np.dot(reference.T, reference)
 
         # get the scaling factor for clean sources