mac m1 usability fix

model/modules.py

@@ -11,7 +11,8 @@
 
 from utils.tools import get_mask_from_lengths, pad
 
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+# device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
 
 
 class VarianceAdaptor(nn.Module):
@@ -77,14 +78,22 @@ def __init__(self, preprocess_config, model_config):
             n_bins, model_config["transformer"]["encoder_hidden"]
         )
 
+    def manual_bucketize(self, input_tensor, boundaries):
+        boundaries = torch.tensor(boundaries).float()
+        expanded_input = input_tensor.unsqueeze(-1)
+        comparison = (expanded_input > boundaries).float()
+        bucket_indices = comparison.sum(-1).long()
+        return bucket_indices
+
+
     def get_pitch_embedding(self, x, target, mask, control):
         prediction = self.pitch_predictor(x, mask)
         if target is not None:
             embedding = self.pitch_embedding(torch.bucketize(target, self.pitch_bins))
         else:
             prediction = prediction * control
             embedding = self.pitch_embedding(
-                torch.bucketize(prediction, self.pitch_bins)
+                self.manual_bucketize(prediction, self.pitch_bins)
             )
         return prediction, embedding
 
@@ -95,7 +104,7 @@ def get_energy_embedding(self, x, target, mask, control):
         else:
             prediction = prediction * control
             embedding = self.energy_embedding(
-                torch.bucketize(prediction, self.energy_bins)
+                self.manual_bucketize(prediction, self.energy_bins)
             )
         return prediction, embedding
 

requirements.txt

@@ -3,15 +3,16 @@ inflect == 4.1.0
 librosa == 0.7.2
 matplotlib == 3.2.2
 numba == 0.48
-numpy == 1.19.0
+# numpy == 1.19.0
 pypinyin==0.39.0
-pyworld == 0.2.10
-PyYAML==5.4.1
-scikit-learn==0.23.2
-scipy == 1.5.0
-soundfile==0.10.3.post1
-tensorboard == 2.2.2
-tgt == 1.4.4
-torch == 1.7.0
-tqdm==4.46.1
-unidecode == 1.1.1
+# pyworld == 0.2.10
+pyworld==0.3.4
+# PyYAML==5.4.1
+scikit-learn
+scipy
+soundfile
+tensorboard
+tgt
+torch
+tqdm
+unidecode

synthesize.py

@@ -14,8 +14,9 @@
 from dataset import TextDataset
 from text import text_to_sequence
 
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-
+# device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
+print('device:', device)
 
 def read_lexicon(lex_path):
     lexicon = {}

utils/model.py

@@ -17,7 +17,7 @@ def get_model(args, configs, device, train=False):
             train_config["path"]["ckpt_path"],
             "{}.pth.tar".format(args.restore_step),
         )
-        ckpt = torch.load(ckpt_path)
+        ckpt = torch.load(ckpt_path, map_location=torch.device('mps'))
         model.load_state_dict(ckpt["model"])
 
     if train:
@@ -55,12 +55,14 @@ def get_vocoder(config, device):
         vocoder.mel2wav.eval()
         vocoder.mel2wav.to(device)
     elif name == "HiFi-GAN":
+        print('device:', device)
         with open("hifigan/config.json", "r") as f:
             config = json.load(f)
         config = hifigan.AttrDict(config)
         vocoder = hifigan.Generator(config)
         if speaker == "LJSpeech":
-            ckpt = torch.load("hifigan/generator_LJSpeech.pth.tar")
+            ckpt = torch.load("hifigan/generator_LJSpeech.pth.tar", 
+                              map_location=torch.device('mps'))
         elif speaker == "universal":
             ckpt = torch.load("hifigan/generator_universal.pth.tar")
         vocoder.load_state_dict(ckpt["generator"])

utils/tools.py

@@ -12,7 +12,8 @@
 matplotlib.use("Agg")
 
 
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+# device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
 
 
 def to_device(data, device):