Update app.py

app.py

@@ -54,13 +54,20 @@ pipe_dict = {
     "language": "english",
 }
 
-title = "# 🐶 VITS"
+title =      """# Explore MMS finetuning
+            ## Or how to access truely multilingual TTS
 
-max_speakers = 15
-
-description = """
+            Massively Multilingual Speech (MMS) models are light-weight, low-latency TTS models based on the [VITS architecture](https://huggingface.co/docs/transformers/model_doc/vits).
+            
+            Meta's [MMS](https://arxiv.org/abs/2305.13516) project, aiming to provide speech technology across a diverse range of languages. You can find more details about the supported languages and their ISO 639-3 codes in the [MMS Language Coverage Overview](https://dl.fbaipublicfiles.com/mms/misc/language_coverage_mms.html),
+    and see all MMS-TTS checkpoints on the Hugging Face Hub: [facebook/mms-tts](https://huggingface.co/models?sort=trending&search=facebook%2Fmms-tts).
+                
+            Coupled with the right data and the right training recipe, you can get an excellent finetuned version of every MMS checkpoints in **20 minutes** with as little as **80 to 150 samples**.            
+            
+            Stay tuned, the training recipe is coming soon!
+            """
 
-"""
+max_speakers = 15
 
 
 # Inference
@@ -104,13 +111,24 @@ def generate_audio(text, model_id, language):
     return out
 
 
+css = """
+#container{
+    margin: 0 auto;
+    max-width: 80rem;
+}
+#intro{
+    max-width: 100%;
+    text-align: center;
+    margin: 0 auto;
+}
+"""
 # Gradio blocks demo    
-with gr.Blocks() as demo_blocks:
-    gr.Markdown(title)
-    gr.Markdown(description)
+with gr.Blocks(css=css) as demo_blocks:
+    gr.Markdown(title, elem_id="intro")
+
     with gr.Row():
         with gr.Column():
-            inp_text = gr.Textbox(label="Input Text", info="What would you like VITS to synthesise?")
+            inp_text = gr.Textbox(label="Input Text", info="What sentence would you like to synthesise?")
             btn = gr.Button("Generate Audio!")
             language = gr.Dropdown(
                 default_model_per_language.keys(),
@@ -120,18 +138,55 @@ with gr.Blocks() as demo_blocks:
             )
             
             model_id = gr.Dropdown(
-                    models_per_language["english"],
+                    models_per_language["spanish"],
                     value="ylacombe/mms-spa-finetuned-chilean-monospeaker", 
                     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)
 
+    gr.Markdown("""
+    ## Datasets and models details
+    
+    For each language, we used 100 to 150 samples of a single speaker to finetune the model.
+    
+    ### Spanish
+    
+    * **Model**: [Spanish MMS TTS](https://huggingface.co/facebook/mms-tts-spa).
+    * **Datasets**:
+        - [Chilean Spanish TTS dataset](https://huggingface.co/datasets/ylacombe/google-chilean-spanish).
+
+    ### Tamil
+    
+    * **Model**: [Tamil MMS TTS](https://huggingface.co/facebook/mms-tts-tam).
+    * **Datasets**:
+        - [Tamil TTS dataset](https://huggingface.co/datasets/ylacombe/google-tamil).
+
+    ### Gujarati
+    
+    * **Model**: [Gujarati MMS TTS](https://huggingface.co/facebook/mms-tts-guj).
+    * **Datasets**:
+        - [Gujarati TTS dataset](https://huggingface.co/datasets/ylacombe/google-gujarati).
+
+    ### Marathi
+    
+    * **Model**: [Marathi MMS TTS](https://huggingface.co/facebook/mms-tts-mar).
+    * **Datasets**:
+        - [Marathi TTS dataset](https://huggingface.co/datasets/ylacombe/google-chilean-marathi).
+    
+    ### English
+    
+    * **Model**: [VITS-ljs](https://huggingface.co/kakao-enterprise/vits-ljs)
+    * **Dataset**: [British Isles Accent](https://huggingface.co/datasets/ylacombe/english_dialects). For each accent, we used 100 to 150 samples of a single speaker to finetune [VITS-ljs](https://huggingface.co/kakao-enterprise/vits-ljs).
+    
+                
+                """) 
+
     language.change(lambda language: gr.Dropdown(
                     models_per_language[language],
                     value=models_per_language[language][0],