support tune f0

app.py

@@ -55,10 +55,10 @@ def tune_f0(initial_f0, i):
     if i == 0:
         return initial_f0
     voiced = initial_f0 > threshold
-    initial_lf0 = torch.log(initial_f0)
+    initial_lf0 = np.log(initial_f0)
     lf0 = initial_lf0 + step * i
-    f0 = torch.exp(lf0)
-    f0 = torch.where(voiced, f0, initial_f0)
+    f0 = np.exp(lf0)
+    f0 = np.where(voiced, f0, initial_f0)
     return f0
 
 # infer
@@ -98,6 +98,7 @@ def convert(tgt_spk, src_wav, f0_shift=0):
 
         f0_mean_tgt = f0_stats[tgt_spk]["mean"]
         f0_mean_tgt = np.array([f0_mean_tgt], dtype=np.float32)[None, :]
+        f0_mean_tgt = tune_f0(f0_mean_tgt, f0_shift)
 
         # src
         wav, sr = librosa.load(src_wav, sr=16000)