Shared look + usage stats

app.py

@@ -7,6 +7,8 @@ from pathlib import Path
 import gradio as gr
 import numpy as np
 import spaces
+
+import mfn_hf  # shared look + usage analytics; import before @spaces.GPU
 import torch
 import torchaudio
 import soundfile as sf
@@ -89,7 +91,7 @@ MODES = {
 }
 
 @spaces.GPU(duration=120)
-def generate_speech(
+def _generate_speech(
     prompt_audio,
     prompt_text,
     target_text,
@@ -148,35 +150,59 @@ def generate_speech(
     return (24000, audio_np)
 
 
+def generate_speech(
+    prompt_audio,
+    prompt_text,
+    target_text,
+    mode,
+    temperature,
+    custom_teacher_steps,
+    custom_teacher_stopping_time,
+    custom_student_start_step,
+    verbose,
+):
+    t0 = time.time()
+    audio = _generate_speech(
+        prompt_audio,
+        prompt_text,
+        target_text,
+        mode,
+        temperature,
+        custom_teacher_steps,
+        custom_teacher_stopping_time,
+        custom_student_start_step,
+        verbose,
+    )
+    return audio, f"Took {time.time() - t0:.1f} s"
+
+
 # Create Gradio interface
 with gr.Blocks(title="DMOSpeech 2 - Zero-Shot TTS") as demo:
-    gr.Markdown(
-        f"""
-    # 🎙️ DMOSpeech 2: Zero-Shot Text-to-Speech
-
-    [GitHub Repo](https://github.com/yl4579/DMOSpeech2)
-    
-    Generate natural speech in any voice with just a short reference audio!     
-    """
+    mfn_hf.header(
+        "DMOSpeech 2",
+        model="yl4579/DMOSpeech2",
+        description="Speak any text in any voice from a short sample of that voice.",
+        flow_fn="/generate_speech",
     )
+    gr.Markdown("Code and paper: [github.com/yl4579/DMOSpeech2](https://github.com/yl4579/DMOSpeech2)")
 
     with gr.Row():
         with gr.Column(scale=1):
             # Reference audio input
             prompt_audio = gr.Audio(
-                label="📎 Reference Audio",
+                label="Reference Audio",
                 type="filepath",
                 sources=["upload", "microphone"],
             )
 
             prompt_text = gr.Textbox(
-                label="📝 Reference Text (leave empty for auto-transcription)",
+                label="Reference Text (leave empty for auto-transcription)",
                 placeholder="The text spoken in the reference audio...",
                 lines=2,
             )
 
             target_text = gr.Textbox(
-                label="✍️ Text to Generate",
+                label="Text to Generate",
                 placeholder="Enter the text you want to synthesize...",
                 lines=4,
             )
@@ -190,12 +216,12 @@ with gr.Blocks(title="DMOSpeech 2 - Zero-Shot TTS") as demo:
                     "Custom",
                 ],
                 value="Teacher-Guided (8 steps)",
-                label="🚀 Generation Mode",
+                label="Generation Mode",
                 info="Choose speed vs quality/diversity tradeoff",
             )
 
             # Advanced settings (collapsible)
-            with gr.Accordion("⚙️ Advanced Settings", open=False):
+            with gr.Accordion("Advanced Settings", open=False):
                 temperature = gr.Slider(
                     minimum=0.0,
                     maximum=2.0,
@@ -240,18 +266,19 @@ with gr.Blocks(title="DMOSpeech 2 - Zero-Shot TTS") as demo:
                     info="Show detailed generation steps",
                 )
 
-            generate_btn = gr.Button("🎵 Generate Speech", variant="primary", size="lg")
+            generate_btn = gr.Button("Generate Speech", variant="primary", size="lg")
 
         with gr.Column(scale=1):
             # Output
             output_audio = gr.Audio(
-                label="🔊 Generated Speech", type="filepath", autoplay=True
+                label="Generated Speech", type="filepath", autoplay=True
             )
+            took = mfn_hf.took()
 
             # Tips
             gr.Markdown(
                 """
-            ### 💡 Quick Tips:
+            ### Tips
             
             - **Auto-transcription**: Leave reference text empty to auto-transcribe
             - **Student Only**: Fastest (4 steps), good quality
@@ -259,7 +286,7 @@ with gr.Blocks(title="DMOSpeech 2 - Zero-Shot TTS") as demo:
             - **High Diversity**: More natural prosody (16 steps)
             - **Custom Mode**: Fine-tune all parameters
             
-            ### 📊 Expected RTF (Real-Time Factor):
+            ### Speed
             - Student Only: ~0.05x (20x faster than real-time)
             - Teacher-Guided: ~0.10x (10x faster)
             - High Diversity: ~0.20x (5x faster)
@@ -280,9 +307,11 @@ with gr.Blocks(title="DMOSpeech 2 - Zero-Shot TTS") as demo:
             custom_student_start_step,
             verbose,
         ],
-        outputs=[output_audio],
+        outputs=[output_audio, took],
     )
 
     mode.change(lambda x: gr.update(visible=x == "Custom"), inputs=[mode], outputs=[custom_settings])
 
+    mfn_hf.footer()
+
 demo.queue().launch()

requirements.txt

@@ -37,4 +37,5 @@ x_transformers>=1.31.14
 # zhconv
 # zhon
 
-cached_path
+cached_path
+mfn_hf @ https://api.mrfake.name/dist/mfn_hf-latest.tar.gz