Upload 6 files

app.py

@@ -39,6 +39,8 @@ languages = [l.value for l in Languages]
 
 
 def tts_fn(
+    model_name,
+    model_path,
     text,
     language,
     reference_audio_path,
@@ -64,7 +66,7 @@ def tts_fn(
             None,
             kata_tone_json_str,
         )
-    assert model_holder.current_model is not None
+    model_holder.load_model_gr(model_name, model_path)
 
     wrong_tone_message = ""
     kata_tone: Optional[list[tuple[str, int]]] = None
@@ -255,7 +257,8 @@ if __name__ == "__main__":
                 text_input = gr.TextArea(label="テキスト", value=initial_text)
 
                 line_split = gr.Checkbox(
-                    label="改行で分けて生成", value=DEFAULT_LINE_SPLIT
+                    label="改行で分けて生成（分けたほうが感情が乗ります）",
+                    value=DEFAULT_LINE_SPLIT,
                 )
                 split_interval = gr.Slider(
                     minimum=0.0,

common/tts_model.py

@@ -209,6 +209,18 @@ class ModelHolder:
             raise ValueError(f"Model `{model_name}` is not found")
         if model_path not in self.model_files_dict[model_name]:
             raise ValueError(f"Model file `{model_path}` is not found")
+        if (
+            self.current_model is not None
+            and self.current_model.model_path == model_path
+        ):
+            # Already loaded
+            speakers = list(self.current_model.spk2id.keys())
+            styles = list(self.current_model.style2id.keys())
+            return (
+                gr.Dropdown(choices=styles, value=styles[0]),
+                gr.Button(interactive=True, value="音声合成"),
+                gr.Dropdown(choices=speakers, value=speakers[0]),
+            )
         self.current_model = Model(
             model_path=model_path,
             config_path=os.path.join(self.root_dir, model_name, "config.json"),