fix: replace music with MusicGen AI-generated tracks

scripts/generate-music-ai.py

@@ -0,0 +1,80 @@
+#!/usr/bin/env python3
+"""Generate game music using Meta's MusicGen AI model."""
+
+import torch
+import soundfile as sf
+import numpy as np
+import os
+
+OUTPUT_DIR = os.path.join(os.path.dirname(__file__), "..", "client", "public", "music")
+
+def generate_tracks():
+    from audiocraft.models import MusicGen
+
+    print("Loading MusicGen medium model...")
+    model = MusicGen.get_pretrained('facebook/musicgen-medium')
+    print("Model loaded!")
+
+    tracks = [
+        {
+            "name": "ambient",
+            "prompt": "dark fantasy ambient music, mysterious dungeon atmosphere, low ethereal drone, subtle reverb chimes, medieval dark orchestral pad, haunting, slow tempo, atmospheric, cinematic",
+            "duration": 30,
+        },
+        {
+            "name": "combat",
+            "prompt": "intense dark fantasy battle music, epic orchestral combat theme, driving war drums, aggressive strings, minor key, brass hits, fast tempo 140bpm, action RPG boss fight music",
+            "duration": 20,
+        },
+        {
+            "name": "victory",
+            "prompt": "triumphant fantasy victory fanfare, heroic orchestral brass, major key celebration, uplifting strings, RPG level complete music, glorious achievement",
+            "duration": 10,
+        },
+        {
+            "name": "defeat",
+            "prompt": "somber dark fantasy defeat theme, melancholic minor key strings, slow mournful cello, ambient sadness, game over music, fading hope",
+            "duration": 8,
+        },
+    ]
+
+    os.makedirs(OUTPUT_DIR, exist_ok=True)
+
+    for track in tracks:
+        print(f"\nGenerating {track['name']} ({track['duration']}s)...")
+        print(f"  Prompt: {track['prompt']}")
+
+        model.set_generation_params(
+            duration=track["duration"],
+            top_k=250,
+            top_p=0.0,
+            temperature=1.0,
+        )
+
+        wav = model.generate([track["prompt"]])
+
+        # wav shape: [batch, channels, samples]
+        audio = wav[0]  # First (only) batch item
+
+        # Normalize to prevent clipping
+        peak = audio.abs().max()
+        if peak > 0:
+            audio = audio / peak * 0.9
+
+        path = os.path.join(OUTPUT_DIR, f"{track['name']}.wav")
+
+        # Remove old file
+        for ext in ['.wav', '.mp3']:
+            old = os.path.join(OUTPUT_DIR, f"{track['name']}{ext}")
+            if os.path.exists(old):
+                os.remove(old)
+
+        audio_np = audio.cpu().numpy().squeeze()
+        sf.write(path, audio_np, model.sample_rate)
+        size_kb = os.path.getsize(path) / 1024
+        print(f"  Saved: {path} ({size_kb:.0f}KB)")
+
+    print("\nAll tracks generated!")
+
+if __name__ == "__main__":
+    generate_tracks()