Update app.py

app.py

@@ -17,93 +17,65 @@ device = _grab_best_device()
 default_model_per_language = {
     "english": "kakao-enterprise/vits-ljs",
     "spanish": "facebook/mms-tts-spa",
-    "tamil": "facebook/mms-tts-tam"
 }
 
 models_per_language = {
     "english": [
-        "ylacombe/vits_ljs_irish_male_monospeaker",
-        "ylacombe/vits_ljs_irish_male_monospeaker_2",
-        "ylacombe/vits_ljs_irish_male_monospeaker_2",
-        "ylacombe/vits_ljs_irish_male_2",
-
-        "ylacombe/vits_ljs_welsh_female_monospeaker",
-        "ylacombe/vits_ljs_welsh_female_monospeaker_2",
-        "ylacombe/vits_ljs_welsh_female_2",
-
-        "ylacombe/vits_ljs_welsh_male_monospeaker",
-        "ylacombe/vits_ljs_welsh_male_monospeaker_2",
-
-        "ylacombe/vits_ljs_scottish_female_monospeaker",
-        "ylacombe/vits_ljs_scottish_female_2",
+        ("Irish Male Speaker", "ylacombe/vits_ljs_irish_male_monospeaker_2"),
+        ("Welsh Female Speaker", "ylacombe/vits_ljs_welsh_female_monospeaker_2"),
+        ("Welsh Male Speaker", "ylacombe/vits_ljs_welsh_male_monospeaker_2"),
+        ("Scottish Female Speaker", "ylacombe/vits_ljs_scottish_female_monospeaker"),
     ],
     "spanish": [
-        "ylacombe/mms-spa-finetuned-chilean-monospeaker-all",
-        "ylacombe/mms-spa-finetuned-chilean-monospeaker",       
+        ("Male Chilean Speaker", "ylacombe/mms-spa-finetuned-chilean-monospeaker"),       
+        ("Female Argentinian Speaker", "ylacombe/mms-spa-finetuned-argentinian-monospeaker"),       
+        ("Male Colombian Speaker", "ylacombe/mms-spa-finetuned-colombian-monospeaker"),       
     ],
-    "tamil": [
-        "ylacombe/mms-tam-finetuned-monospeaker-all",
-        "ylacombe/mms-tam-finetuned-monospeaker",
-    ]
 }
 
-HUB_PATH = "ylacombe/vits_ljs_welsh_female_monospeaker"
-
-
 pipe_dict = {
-    "current_model": "ylacombe/vits_ljs_welsh_female_monospeaker",
-    "pipe":  pipeline("text-to-speech", model=HUB_PATH, device=0),
+    "pipe":  [pipeline("text-to-speech", model=l[1], device=0) for l in models_per_language["english"]],
     "original_pipe": pipeline("text-to-speech", model=default_model_per_language["english"], device=0),
     "language": "english",
 }
 
-title = "# 🐶 VITS"
-
-max_speakers = 15
+title = "# VITS"
 
 description = """
-
+TODO
 """
 
+max_speakers = 15
 
 # Inference
-def generate_audio(text, model_id, language):
+def generate_audio(text, language):
 
     if pipe_dict["language"] != language:
-        gr.Warning(f"Language has changed - loading new default model: {default_model_per_language[language]}")
+        gr.Warning(f"Language has changed - loading corresponding models: {default_model_per_language[language]}")
         pipe_dict["language"] = language
         pipe_dict["original_pipe"] = pipeline("text-to-speech", model=default_model_per_language[language], device=0)
     
-    if pipe_dict["current_model"] != model_id:
-        gr.Warning("Model has changed - loading new model")
-        pipe_dict["pipe"] = pipeline("text-to-speech", model=model_id, device=0)
-        pipe_dict["current_model"] = model_id
+        pipe_dict["pipe"] = [pipeline("text-to-speech", model=l[1], device=0) for l in models_per_language["english"]]
 
     num_speakers = pipe_dict["pipe"].model.config.num_speakers
 
     out = []
     # first generate original model result
     output = pipe_dict["original_pipe"](text)
-    output =  gr.Audio(value = (output["sampling_rate"], output["audio"].squeeze()), type="numpy", autoplay=False, label=f"Non finetuned model prediction {default_model_per_language[language]}", show_label=True,
+    output =  gr.Audio(value = (output["sampling_rate"], output["audio"].squeeze()), type="numpy", autoplay=False, label=f"Prediction from the original checkpoint {default_model_per_language[language]}", show_label=True,
                                visible=True)
     out.append(output)
     
     
-    if num_speakers>1:
-        for i in range(min(num_speakers, max_speakers - 1)):
-            forward_params = {"speaker_id": i}
-            output = pipe_dict["pipe"](text, forward_params=forward_params)
-            
-            output =  gr.Audio(value = (output["sampling_rate"], output["audio"].squeeze()), type="numpy", autoplay=False, label=f"Generated Audio - speaker {i}", show_label=True,
-                               visible=True)
-            out.append(output)
-        out.extend([gr.Audio(visible=False)]*(max_speakers-num_speakers))
-    else:
-        output = pipe_dict["pipe"](text)
-        output =  gr.Audio(value = (output["sampling_rate"], output["audio"].squeeze()), type="numpy", autoplay=False, label="Generated Audio - Mono speaker", show_label=True,
-                               visible=True)
+    for i in range(min(len(pipe_dict["pipe"]), max_speakers - 1)):
+        
+        output = pipe_dict["pipe"][i](text)
+        
+        output =  gr.Audio(value = (output["sampling_rate"], output["audio"].squeeze()), type="numpy", autoplay=False, label=f"Finetuned {models_per_language[language][i][0]}", show_label=True,
+                           visible=True)
         out.append(output)
-        out.extend([gr.Audio(visible=False)]*(max_speakers-2))
+    
+    out.extend([gr.Audio(visible=False)]*(max_speakers-(len(out))))
     return out
 
 
@@ -122,30 +94,13 @@ with gr.Blocks() as demo_blocks:
                 info = "Language that you want to test"
             )
             
-            model_id = gr.Dropdown(
-                    models_per_language["english"],
-                    value="ylacombe/vits_ljs_welsh_female_monospeaker_2", 
-                    label="Model", 
-                    info="Model you want to test",
-                    )
-    
         with gr.Column():
             outputs = []
             for i in range(max_speakers):
                 out_audio = gr.Audio(type="numpy", autoplay=False, label=f"Generated Audio - speaker {i}", show_label=True, visible=False)
                 outputs.append(out_audio)
-
-    language.change(lambda language: gr.Dropdown(
-                    models_per_language[language],
-                    value=models_per_language[language][0], 
-                    label="Model", 
-                    info="Model you want to test",
-                    ),
-                    language,
-                    model_id
-                   )
     
-    btn.click(generate_audio, [inp_text, model_id, language], outputs)
+    btn.click(generate_audio, [inp_text, language], outputs)
     
 
 demo_blocks.queue().launch()