Create app.py

app.py

@@ -0,0 +1,140 @@
+#!/usr/bin/env python3
+"""Music Transcriber - Gradio Web App for Hugging Face Spaces"""
+
+import gradio as gr
+from pathlib import Path
+import tempfile
+import os
+
+# Redirect caches to /tmp for HF Spaces
+os.environ['HF_HOME'] = '/tmp/hf_cache'
+os.environ['TORCH_HOME'] = '/tmp/hf_cache/torch'
+os.environ['TRANSFORMERS_CACHE'] = '/tmp/hf_cache'
+
+def transcribe_audio(audio_file, model_size, instruments):
+    """Transcribe audio to MIDI with progress tracking."""
+    if audio_file is None:
+        return None, "❌ Please upload an audio file first.", 0
+    
+    from muscriptor import TranscriptionModel
+    from muscriptor.events import ProgressEvent
+    
+    # Parse instrument filter
+    instrument_group = None
+    instrument_names = []
+    if instruments:
+        try:
+            from muscriptor.tokenizer.mt3 import MT3_FULL_PLUS_GROUP_NAMES
+            names = [n.strip().lower() for n in instruments.split(",")]
+            valid_names = [n for n in names if n in MT3_FULL_PLUS_GROUP_NAMES]
+            instrument_group = " ".join(str(MT3_FULL_PLUS_GROUP_NAMES[n]) for n in valid_names)
+            instrument_names = valid_names
+        except Exception:
+            pass
+    
+    try:
+        model = TranscriptionModel.load_model(model_size)
+    except Exception as e:
+        return None, f"❌ Model load failed: {e}", 0
+    
+    output_path = Path(tempfile.mktemp(suffix='.mid'))
+    
+    try:
+        note_events = []
+        progress_status = "Starting..."
+        
+        for event in model.transcribe(str(audio_file), instrument_group=instrument_group):
+            if isinstance(event, ProgressEvent):
+                progress_status = f"🎹 Processing: {int(event.fraction * 100)}%"
+            elif hasattr(event, 'note_id'):
+                note_events.append(event)
+        
+        midi_bytes = model.transcribe_to_midi(str(audio_file), instrument_group=instrument_group)
+        output_path.write_bytes(midi_bytes)
+        
+        msg = f"✅ Complete!\n🎼 {len(note_events)} notes detected\n📁 MIDI ready for download"
+        if instrument_names:
+            msg += f"\n🔍 Filtered: {', '.join(instrument_names)}"
+        
+        return str(output_path), msg, len(note_events)
+        
+    except Exception as e:
+        if output_path.exists():
+            output_path.unlink()
+        return None, f"❌ Error: {e}", 0
+
+def list_instruments():
+    """Return list of available instruments."""
+    try:
+        from muscriptor.tokenizer.mt3 import MT3_FULL_PLUS_GROUP_NAMES
+        instruments = sorted(MT3_FULL_PLUS_GROUP_NAMES.keys())
+        return "\n".join(f"• {name}" for name in instruments)
+    except Exception:
+        return "Could not load instrument list. Install muscriptor first."
+
+# Build Gradio Interface
+with gr.Blocks(
+    title="🎵 Music Transcriber",
+    theme=gr.themes.Soft(primary_hue="purple"),
+    css="""
+        .gradio-container { max-width: 900px !important; }
+        .note-count { font-size: 2rem; font-weight: bold; color: #8b5cf6; text-align: center; padding: 1rem; }
+        .instructions { font-size: 0.9rem; color: #6b7280; }
+    """
+) as demo:
+    
+    gr.Markdown("# 🎵 Music Transcriber\nTurn audio into MIDI using AI. Upload a song, select model size, and get your transcription!")
+    
+    with gr.Row():
+        with gr.Column(scale=1):
+            audio_input = gr.Audio(
+                type="filepath",
+                label="Upload Audio (MP3, WAV, FLAC, OGG, M4A)",
+                interactive=True
+            )
+            
+            with gr.Row():
+                model_select = gr.Radio(
+                    choices=[("Small (~100M) - Fastest ⚡", "small"),
+                            ("Medium (~300M) - Balanced 👍", "medium"),
+                            ("Large (~1.3B) - Best Quality 🌟", "large")],
+                    value="medium",
+                    label="Model Size"
+                )
+            
+            instrument_input = gr.Textbox(
+                label="Filter Instruments (optional)",
+                placeholder="e.g., acoustic_piano, acoustic_guitar, electric_bass",
+                info="Comma-separated instrument names. Leave blank for all instruments."
+            )
+            
+            transcribe_btn = gr.Button("🎹 Transcribe Audio", variant="primary", size="lg")
+        
+        with gr.Column(scale=1):
+            status_output = gr.Textbox(label="Status", lines=4)
+            note_count_output = gr.HTML('<div class="note-count">0</div>')
+            midi_output = gr.File(label="Download MIDI", visible=False)
+    
+    # List instruments accordion
+    with gr.Accordion("📋 View Available Instruments", open=False):
+        gr.Markdown("```")
+        inst_list = gr.Markdown(label="Instruments")
+        gr.Markdown("```")
+    
+    # Button callbacks
+    transcribe_btn.click(
+        fn=transcribe_audio,
+        inputs=[audio_input, model_select, instrument_input],
+        outputs=[midi_output, status_output, note_count_output]
+    ).then(
+        lambda path, notes: (gr.update(visible=bool(path)), f"<div class='note-count'>{notes}</div>"),
+        inputs=[midi_output, note_count_output],
+        outputs=[midi_output, note_count_output]
+    )
+    
+    # Initialize instrument list
+    gr.Markdown("Click the accordion above to see all available instruments.")
+    inst_list.value = list_instruments()
+
+if __name__ == "__main__":
+    demo.launch()