|
|
@@ -12,7 +12,7 @@ We allow annotations to be compared between different recordings only because al
|
|
|
"""
|
|
|
speakers = ['CHI', 'OCH', 'FEM', 'MAL']
|
|
|
speakers_plot = speakers
|
|
|
-speakers_plot.append("None")
|
|
|
+#speakers_plot.append("None")
|
|
|
speakers_plot = [""] + speakers_plot
|
|
|
|
|
|
project = ChildProject('.')
|
|
|
@@ -24,26 +24,7 @@ anns = pd.DataFrame.copy(am.annotations)
|
|
|
human_ann = anns[anns['set'] == 'eaf_cha']
|
|
|
vtc_ann = anns[anns['set'] == 'vtc']
|
|
|
|
|
|
-def initiate_plt():
|
|
|
- fig, ax = plt.subplots(figsize=(7.5, 7.5))
|
|
|
- ax.set_xticklabels(speakers)
|
|
|
- ax.set_yticklabels(speakers)
|
|
|
- ax.xaxis.set_label_position("top")
|
|
|
- plt.xlabel('HUMAN', fontsize=18)
|
|
|
- plt.ylabel('VTC', fontsize=18)
|
|
|
- plt.title('Confusion Matrix', fontsize=18)
|
|
|
- return ax
|
|
|
-
|
|
|
-#ax = initiate_plt()
|
|
|
-
|
|
|
-def build_matrix(array, output):
|
|
|
- fig = plt.figure(figsize=(12, 6.75))
|
|
|
-
|
|
|
- ax.matshow(array, cmap=plt.cm.Blues, alpha=0.3)
|
|
|
- for i in range(array.shape[0]):
|
|
|
- for j in range(array.shape[1]):
|
|
|
- ax.text(x=j, y=i,s=round(array[i, j],3), va='center', ha='center', size='xx-large')
|
|
|
- plt.savefig(output)
|
|
|
+precisions = pd.DataFrame()
|
|
|
|
|
|
for i, row in vtc_ann.iterrows():
|
|
|
audio_rec = row['recording_filename']
|
|
|
@@ -62,12 +43,11 @@ for i, row in vtc_ann.iterrows():
|
|
|
eaf_cha = segments_to_grid(segments[segments['set'] == 'eaf_cha'], 0, segments['segment_offset'].max(), 100, 'speaker_type', speakers)
|
|
|
|
|
|
confusion_counts = conf_matrix(vtc, eaf_cha)
|
|
|
-
|
|
|
- #build_matrix(confusion_counts, 'extra/reliability/{}.png'.format(audio_rec),ax)
|
|
|
-
|
|
|
- #normalized = confusion_counts/(np.sum(vtc, axis = 0)[:,None])
|
|
|
-
|
|
|
- #build_matrix(normalized,'extra/reliability/{}_normalized.png'.format(audio_rec),ax)
|
|
|
+
|
|
|
+ total = np.sum(confusion_counts)
|
|
|
+ precision = np.trace(confusion_counts) / total if total != 0 else 0
|
|
|
+
|
|
|
+ precisions = pd.concat([precisions, pd.DataFrame(data = {'recording_filename': [audio_rec],'precision': [precision]})])
|
|
|
|
|
|
normalized = confusion_counts
|
|
|
|
|
|
@@ -101,4 +81,7 @@ for i, row in vtc_ann.iterrows():
|
|
|
plt.title('Confusion Matrix', fontsize=18)
|
|
|
plt.savefig('extra/reliability/{}_normalized.png'.format(audio_rec))
|
|
|
|
|
|
- plt.close()
|
|
|
+ plt.close()
|
|
|
+
|
|
|
+print(precisions)
|
|
|
+precisions.to_csv('extra/reliability/precisions.csv',index=False)
|
