Update app.py

app.py

@@ -143,6 +143,26 @@ def get_supported_languages_display() -> str:
 {line2}
 """
 
+def format_for_singing(lyrics: str) -> str:
+    return f"""
+You are a playful children's song singer.
+Do NOT speak normally.
+Perform this rhythmically and melodically like a song.
+
+Rules:
+- Stretch vowels
+- Follow rhythm
+- Pause between lines
+- Raise pitch on questions
+- Sound playful and musical
+
+Start with a soft humming intro:
+hmm-hmm-hmm ♪
+
+Lyrics (sing line by line):
+
+{lyrics}
+"""
 
 DEVICE = "cpu"
 MODEL = None
@@ -206,6 +226,8 @@ def resolve_audio_prompt(language_id: str, provided_path: str | None) -> str | N
 
 def generate_tts_audio(
     text_input: str,
+    lyrics_input: str,
+    mode: str,
     language_id: str,
     audio_prompt_path_input: str = None,
     exaggeration_input: float = 0.5,
@@ -232,89 +254,109 @@ def generate_tts_audio(
     if chosen_prompt:
         generate_kwargs["audio_prompt_path"] = chosen_prompt
 
+    # 🔀 Choose Speak vs Sing text
+    if mode == "Sing 🎵" and lyrics_input.strip():
+        final_text = format_for_singing(lyrics_input)
+    else:
+        final_text = text_input
+
     # 🔒 CPU-safe inference
     with torch.no_grad():
         wav = current_model.generate(
-            text_input[:300],
+            final_text[:300],
             language_id=language_id,
             **generate_kwargs
         )
 
-    # Ensure CPU numpy conversion
     wav = wav.squeeze(0).detach().cpu().numpy()
+    return current_model.sr, wav
 
-    return (current_model.sr, wav)
 
 with gr.Blocks() as demo:
     gr.Markdown(
         """
         # Chatterbox Multilingual Demo
-        Generate high-quality multilingual speech from text with reference audio styling, supporting 23 languages.
-        
-        For a hosted version of Chatterbox Multilingual and for finetuning, please visit [resemble.ai](https://app.resemble.ai)
+        Generate high-quality multilingual speech from text or lyrics (sing mode).
         """
     )
-    
-    # Display supported languages
+
     gr.Markdown(get_supported_languages_display())
+
     with gr.Row():
         with gr.Column():
             initial_lang = "hi"
+
+            mode = gr.Radio(
+                choices=["Speak 🗣️", "Sing 🎵"],
+                value="Speak 🗣️",
+                label="Output Mode"
+            )
+
             text = gr.Textbox(
                 value=default_text_for_ui(initial_lang),
-                label="Text to synthesize (max chars 300)",
-                max_lines=5
+                label="Text (Speak mode)",
+                max_lines=4
             )
-            
+
+            lyrics = gr.Textbox(
+                label="Lyrics (Sing mode)",
+                placeholder="Paste lyrics here (one line per verse)",
+                max_lines=10
+            )
+
             language_id = gr.Dropdown(
                 choices=list(ChatterboxMultilingualTTS.get_supported_languages().keys()),
                 value=initial_lang,
-                label="Language",
-                info="Select the language for text-to-speech synthesis"
+                label="Language"
             )
-            
+
             ref_wav = gr.Audio(
                 sources=["upload", "microphone"],
                 type="filepath",
-                label="Reference Audio File (Optional)",
+                label="Reference Audio (Optional)",
                 value=default_audio_for_ui(initial_lang)
             )
-            
-            gr.Markdown(
-                "💡 **Note**: Ensure that the reference clip matches the specified language tag. Otherwise, language transfer outputs may inherit the accent of the reference clip's language. To mitigate this, set the CFG weight to 0.",
-                elem_classes=["audio-note"]
-            )
-            
+
             exaggeration = gr.Slider(
-                0.25, 2, step=.05, label="Exaggeration (Neutral = 0.5, extreme values can be unstable)", value=.5
+                0.25, 2, step=0.05,
+                label="Exaggeration",
+                value=0.5
             )
+
             cfg_weight = gr.Slider(
-                0.2, 1, step=.05, label="CFG/Pace", value=0.5
+                0.2, 1, step=0.05,
+                label="CFG / Pace",
+                value=0.5
             )
 
             with gr.Accordion("More options", open=False):
-                seed_num = gr.Number(value=0, label="Random seed (0 for random)")
-                temp = gr.Slider(0.05, 5, step=.05, label="Temperature", value=.8)
+                seed_num = gr.Number(value=0, label="Random seed (0 = random)")
+                temp = gr.Slider(0.05, 5, step=0.05, label="Temperature", value=0.8)
 
             run_btn = gr.Button("Generate", variant="primary")
 
         with gr.Column():
             audio_output = gr.Audio(label="Output Audio")
 
-        def on_language_change(lang, current_ref, current_text):
-            return default_audio_for_ui(lang), default_text_for_ui(lang)
-
-        language_id.change(
-            fn=on_language_change,
-            inputs=[language_id, ref_wav, text],
-            outputs=[ref_wav, text],
-            show_progress=False
-        )
+    # 🎛️ Auto-tune sliders for Sing mode
+    def on_mode_change(mode):
+        if mode == "Sing 🎵":
+            return 1.25, 1.0, 0.45
+        return 0.5, 0.8, 0.5
+
+    mode.change(
+        fn=on_mode_change,
+        inputs=mode,
+        outputs=[exaggeration, temp, cfg_weight],
+        show_progress=False
+    )
 
     run_btn.click(
         fn=generate_tts_audio,
         inputs=[
             text,
+            lyrics,
+            mode,
             language_id,
             ref_wav,
             exaggeration,
@@ -326,3 +368,5 @@ with gr.Blocks() as demo:
     )
 
 demo.launch(mcp_server=True)
+
+