Add max steps control in toolbox

c-lins · Nov 6, 2021 · 80aaf32 · 80aaf32
1 parent c396792
commit 80aaf32
Show file tree

Hide file tree

Showing 4 changed files with 17 additions and 4 deletions.
diff --git a/synthesizer/inference.py b/synthesizer/inference.py
@@ -70,7 +70,7 @@ def load(self):
 
     def synthesize_spectrograms(self, texts: List[str],
                                 embeddings: Union[np.ndarray, List[np.ndarray]],
-                                return_alignments=False, style_idx=0, min_stop_token=5):
+                                return_alignments=False, style_idx=0, min_stop_token=5, steps=2000):
         """
         Synthesizes mel spectrograms from texts and speaker embeddings.
 
@@ -125,7 +125,7 @@ def synthesize_spectrograms(self, texts: List[str],
             speaker_embeddings = torch.tensor(speaker_embeds).float().to(self.device)
 
             # Inference
-            _, mels, alignments = self._model.generate(chars, speaker_embeddings, style_idx=style_idx, min_stop_token=min_stop_token)
+            _, mels, alignments = self._model.generate(chars, speaker_embeddings, style_idx=style_idx, min_stop_token=min_stop_token, steps=steps)
             mels = mels.detach().cpu().numpy()
             for m in mels:
                 # Trim silence from end of each spectrogram

diff --git a/synthesizer/models/tacotron.py b/synthesizer/models/tacotron.py
@@ -420,7 +420,7 @@ def forward(self, texts, mels, speaker_embedding):
 
         return mel_outputs, linear, attn_scores, stop_outputs
 
-    def generate(self, x, speaker_embedding=None, steps=200, style_idx=0, min_stop_token=5):
+    def generate(self, x, speaker_embedding=None, steps=2000, style_idx=0, min_stop_token=5):
         self.eval()
         device = next(self.parameters()).device  # use same device as parameters
 

diff --git a/toolbox/__init__.py b/toolbox/__init__.py
@@ -235,7 +235,7 @@ def synthesize(self):
         embed = self.ui.selected_utterance.embed
         embeds = [embed] * len(texts)
         min_token = int(self.ui.token_slider.value())
-        specs = self.synthesizer.synthesize_spectrograms(texts, embeds, style_idx=int(self.ui.style_slider.value()), min_stop_token=min_token)
+        specs = self.synthesizer.synthesize_spectrograms(texts, embeds, style_idx=int(self.ui.style_slider.value()), min_stop_token=min_token, steps=int(self.ui.length_slider.value())*200)
         breaks = [spec.shape[1] for spec in specs]
         spec = np.concatenate(specs, axis=1)
 

diff --git a/toolbox/ui.py b/toolbox/ui.py
@@ -618,6 +618,19 @@ def __init__(self):
         layout_seed.addWidget(self.token_value_label, 2, 1)
         layout_seed.addWidget(self.token_slider, 2, 3)
 
+        self.length_slider = QSlider(Qt.Horizontal)
+        self.length_slider.setTickInterval(1)
+        self.length_slider.setFocusPolicy(Qt.NoFocus)
+        self.length_slider.setSingleStep(1)
+        self.length_slider.setRange(1, 10)
+        self.length_value_label = QLabel("2")
+        self.length_slider.setValue(2)
+        layout_seed.addWidget(QLabel("MaxLength(最大句长):"), 3, 0)
+
+        self.length_slider.valueChanged.connect(lambda s: self.length_value_label.setNum(s))
+        layout_seed.addWidget(self.length_value_label, 3, 1)
+        layout_seed.addWidget(self.length_slider, 3, 3)
+
         gen_layout.addLayout(layout_seed)
 
         self.loading_bar = QProgressBar()