superres

2 files changed, 22 insertions, 10 deletions

diff --git a/NOTES b/NOTES
index 7aed73c..762df82 100644
--- a/NOTES
+++ b/NOTES
@@ -30,8 +30,8 @@ parser.add_argument('--enable_overwrite', action='store_true')
 # TRAIN VOICE CONVERSION NETWORKS
 
 /home/spawn/.virtualenv/yukarin/bin/python train.py \
-  ./20180719133105_2_holly_2_mat.json \
-  ./data/mat-holly-24000/conversion-net/pp-el8-holly-2-mat
+  ./20180719133105.3.mat_2_holly.json \
+  ./data/mat-holly-24000/conversion-net/pp-el0-holly-2-mat
 
 # OPTIONAL: TRAIN ON MULTIPLE GPUS USING RECIPE.JSON (RUNS MULTIPLE SCREENS)
 
@@ -106,11 +106,20 @@ parser.add_argument('-g', '--gpu', type=int)
 
 
 
+mkdir all
+for i in *.wav
+do
+  dir=${i%.*}
+  ../br_misc/split.sh $i 12 $dir
+  cd $dir
+  for j in *.wav
+  do
+    mv "$j" "${dir}_${j}"
+  done
+  cd ..
+  mv $i/* all
+done
 
-
-
-
-
-
-
+cd output/net/
+mkdir concat
 
diff --git a/become_yukarin/acoustic_converter.py b/become_yukarin/acoustic_converter.py
index 62eacff..706c6ac 100644
--- a/become_yukarin/acoustic_converter.py
+++ b/become_yukarin/acoustic_converter.py
@@ -122,17 +122,20 @@ class AcousticConverter(object):
     def convert_from_audio_path(self, path: Path, out_sampling_rate: Optional[int] = None):
         wave = self._wave_process(str(path), test=True)
         feature = self._feature_process(wave, test=True)
-        return self.convert_from_feature(feature, out_sampling_rate)
+        f0 = feature.f0
+        return self.convert_from_feature(feature, out_sampling_rate, f0)
 
     def convert_from_feature_path(self, path: Path, out_sampling_rate: Optional[int] = None):
         feature = self._acoustic_feature_load_process(path, test=True)
         return self.convert_from_feature(feature, out_sampling_rate)
 
-    def convert_from_feature(self, input: AcousticFeature, out_sampling_rate: Optional[int] = None):
+    def convert_from_feature(self, input: AcousticFeature, out_sampling_rate: Optional[int] = None, f0=None):
         if out_sampling_rate is None:
             out_sampling_rate = self.config.dataset.param.voice_param.sample_rate
 
         out = self.convert_to_feature(input=input, out_sampling_rate=out_sampling_rate)
+        if f0 is not None:
+            out.f0 = f0
         out = pyworld.synthesize(
             f0=out.f0.ravel(),
             spectrogram=out.spectrogram,