Upload voxcpm2_vps.py

voxcpm2_vps.py

@@ -6,13 +6,14 @@ Optimized for: Hostinger KVM 4 (4 vCPU, 8GB RAM) or similar VPS
 ⚠️  MEMORY WARNING: VoxCPM2 (2B params) needs ~8-9GB RAM in float32 on CPU.
     An 8GB VPS will need swap space. This script auto-checks and warns.
 
-    If OOM persists, the script auto-falls back to VoxCPM-0.5B (much smaller).
+    If OOM persists, use VoxCPM-0.5B (much smaller).
 
 Features:
 - Auto memory check with swap creation guide
 - Auto float32 conversion for CPU (fixes bfloat16 crash)
-- Graceful fallback to VoxCPM-0.5B if RAM is insufficient
 - All capabilities: TTS, Voice Design, Voice Cloning, Multilingual, Streaming
+- NEW: Multilingual Voice Cloning (clone voice + speak any language)
+- NEW: Multilingual Voice Design (design voice + speak any language)
 """
 
 import os
@@ -20,7 +21,6 @@ import sys
 import time
 import argparse
 import warnings
-import subprocess
 
 warnings.filterwarnings("ignore")
 
@@ -61,21 +61,6 @@ def check_memory():
         return True
 
 
-def create_swap_guide():
-    """Print swap creation commands."""
-    print("""
-╔══════════════════════════════════════════════════════════════════╗
-║  HOW TO ADD SWAP ON YOUR VPS (run these via SSH)               ║
-╠══════════════════════════════════════════════════════════════════╣
-║  sudo fallocate -l 8G /swapfile                                  ║
-║  sudo chmod 600 /swapfile                                        ║
-║  sudo mkswap /swapfile                                           ║
-║  sudo swapon /swapfile                                           ║
-║  echo '/swapfile none swap sw 0 0' | sudo tee -a /etc/fstab      ║
-╚══════════════════════════════════════════════════════════════════╝
-""")
-
-
 def load_model(model_id="openbmb/VoxCPM2"):
     """Load VoxCPM model optimized for VPS CPU."""
     from voxcpm import VoxCPM
@@ -177,10 +162,55 @@ def generate_multilingual(model, texts_dict, timesteps=10):
     return results
 
 
+# NEW: Multilingual Voice Cloning — clone a voice and speak ANY language!
+def generate_multilingual_clone(model, text, reference_wav, output_path="output_multi_clone.wav", timesteps=15):
+    """Clone a voice and speak in ANY language!"""
+    import soundfile as sf
+    if not os.path.exists(reference_wav):
+        print(f"[ERROR] Reference not found: {reference_wav}")
+        return None
+
+    print(f"\n[MODE] Multilingual Voice Cloning")
+    print(f"[TEXT] {text[:80]}...")
+    print(f"[REF]  {reference_wav}")
+    print("[INFO] The cloned voice will speak in whatever language you typed!")
+    
+    start = time.time()
+    wav = model.generate(text=text, reference_wav_path=reference_wav, cfg_value=2.0,
+                         inference_timesteps=timesteps, max_len=4096)
+    elapsed = time.time() - start
+
+    sf.write(output_path, wav, model.tts_model.sample_rate)
+    duration = len(wav) / model.tts_model.sample_rate
+    print(f"[OK] Saved: {output_path} ({duration:.1f}s audio, {elapsed:.1f}s generation)")
+    print("[INFO] ✅ The cloned voice is speaking the language of your text!")
+    return output_path
+
+
+# NEW: Multilingual Voice Design — design a voice and speak ANY language!
+def generate_multilingual_design(model, description, text, output_path="output_multi_design.wav", timesteps=12):
+    """Design a voice and speak in ANY language!"""
+    import soundfile as sf
+    full_text = f"({description}) {text}"
+    print(f"\n[MODE] Multilingual Voice Design")
+    print(f"[DESC] {description}")
+    print(f"[TEXT] {text[:80]}...")
+    print("[INFO] The designed voice will speak in whatever language you typed!")
+    
+    start = time.time()
+    wav = model.generate(text=full_text, cfg_value=2.0, inference_timesteps=timesteps, max_len=4096)
+    elapsed = time.time() - start
+
+    sf.write(output_path, wav, model.tts_model.sample_rate)
+    duration = len(wav) / model.tts_model.sample_rate
+    print(f"[OK] Saved: {output_path} ({duration:.1f}s audio, {elapsed:.1f}s generation)")
+    print("[INFO] ✅ The designed voice is speaking the language of your text!")
+    return output_path
+
+
 def run_all_demos(model_id="openbmb/VoxCPM2"):
     mem_ok = check_memory()
     if not mem_ok:
-        create_swap_guide()
         response = input("\nContinue anyway? (y/n): ").strip().lower()
         if response != 'y':
             print("[INFO] Exiting. Add swap or use --model openbmb/VoxCPM-0.5B")
@@ -192,10 +222,14 @@ def run_all_demos(model_id="openbmb/VoxCPM2"):
     print("RUNNING ALL CAPABILITY DEMOS")
     print("=" * 60)
 
+    # 1. Basic TTS
     generate_basic_tts(model, "Hello! This is VoxCPM2 running on a VPS.", "demo_01_basic.wav", 10)
+    
+    # 2. Voice Design
     generate_voice_design(model, "A warm professional female narrator",
                           "Welcome to today's presentation.", "demo_02_design.wav", 12)
 
+    # 3. Multilingual
     texts = {
         "english": "The quick brown fox jumps over the lazy dog.",
         "chinese": "你好，这是VoxCPM2语音合成模型。",
@@ -203,11 +237,23 @@ def run_all_demos(model_id="openbmb/VoxCPM2"):
     }
     generate_multilingual(model, texts, 10)
 
+    # 4. Voice Clone (if reference exists)
     if os.path.exists("my_voice.wav"):
         generate_voice_clone(model, "This is my cloned voice.", "my_voice.wav", "demo_05_clone.wav", 15)
+        
+        # 5. NEW: Multilingual Clone — clone voice, speak Chinese!
+        generate_multilingual_clone(model, 
+            "你好！这是我克隆的声音在说中文。非常自然和流畅！",
+            "my_voice.wav", "demo_06_multi_clone.wav", 15)
     else:
         print("\n[SKIP] Place 'my_voice.wav' to test voice cloning")
 
+    # 6. NEW: Multilingual Design — design voice, speak Chinese!
+    generate_multilingual_design(model, 
+        "A warm elderly British woman, soft and caring",
+        "你好！这是一个温暖的英国老太太的声音在说中文。",
+        "demo_07_multi_design.wav", 12)
+
     print("\n" + "=" * 60)
     print("ALL DEMOS COMPLETE!")
     print("=" * 60)
@@ -233,12 +279,18 @@ Examples:
 
   # Voice clone
   python voxcpm2_vps.py --mode clone --text "Hi" --reference my_voice.wav
+
+  # Multilingual clone (NEW!) — clone voice + speak any language
+  python voxcpm2_vps.py --mode multi-clone --text "你好！" --reference my_voice.wav
+
+  # Multilingual design (NEW!) — design voice + speak any language
+  python voxcpm2_vps.py --mode multi-design --description "warm elderly woman" --text "¡Hola!"
         """
     )
     parser.add_argument("--model", type=str, default="openbmb/VoxCPM2",
                         help="Model ID: openbmb/VoxCPM2 (2B) or openbmb/VoxCPM-0.5B (smaller)")
     parser.add_argument("--demo", action="store_true", help="Run all capability demos")
-    parser.add_argument("--mode", choices=["basic", "design", "clone", "multilingual"],
+    parser.add_argument("--mode", choices=["basic", "design", "clone", "multilingual", "multi-clone", "multi-design"],
                         default="basic", help="Generation mode")
     parser.add_argument("--text", type=str, default="Hello from VoxCPM2 on VPS!", help="Text to synthesize")
     parser.add_argument("--description", type=str, default="A warm professional voice", help="Voice design description")
@@ -274,6 +326,13 @@ Examples:
             "spanish": "¡Hola! Esta es síntesis de voz.",
         }
         generate_multilingual(model, texts, args.timesteps)
+    elif args.mode == "multi-clone":
+        if not args.reference:
+            print("[ERROR] --reference required for multi-clone mode")
+            sys.exit(1)
+        generate_multilingual_clone(model, args.text, args.reference, args.output, args.timesteps)
+    elif args.mode == "multi-design":
+        generate_multilingual_design(model, args.description, args.text, args.output, args.timesteps)
 
     print(f"\n[OK] Done! Output: {args.output}")