compare_models_english.py

#\!/usr/bin/env python3
import os
import time

print("=== Comparação: CosyVoice-300M vs CosyVoice2-0.5B ===")

# Texto de teste em inglês
test_text = "Hello\! This is a comparison test between CosyVoice models. We are testing English pronunciation quality."

# Teste com modelo atual (300M)
print("\n1. Testando modelo CosyVoice-300M-direct...")
try:
    from cosyvoice.cli.cosyvoice import CosyVoice
    import torchaudio
    
    model1 = CosyVoice('pretrained_models/CosyVoice-300M-direct', load_jit=False, load_trt=False, fp16=False)
    
    start = time.time()
    for i, j in enumerate(model1.inference_zero_shot(test_text, "Testing speech synthesis.", None, stream=False)):
        torchaudio.save('english_300m.wav', j['tts_speech'], model1.sample_rate)
        break
    print(f"✅ Gerado em {time.time()-start:.1f}s - Arquivo: english_300m.wav")
    
except Exception as e:
    print(f"❌ Erro no modelo 300M: {e}")

# Teste com novo modelo (0.5B) - quando estiver disponível
print("\n2. Modelo CosyVoice2-0.5B...")
if os.path.exists('pretrained_models/CosyVoice2-0.5B/llm.pt'):
    print("✅ Modelo baixado\! Pronto para testar.")
else:
    print("⏳ Ainda baixando... Execute novamente após o download.")