final_tts_test.py

#!/usr/bin/env python3
import sys
sys.path.append('third_party/Matcha-TTS')

print("=== CosyVoice TTS Final Test ===\n")

try:
    from cosyvoice.cli.cosyvoice import CosyVoice
    from cosyvoice.utils.file_utils import load_wav
    import torchaudio
    import os
    
    # Use the complete model
    model_path = 'pretrained_models/CosyVoice-300M-direct'
    
    print(f"Loading model from: {model_path}")
    cosyvoice = CosyVoice(model_path, load_jit=False, load_trt=False, fp16=False)
    print("✓ Model loaded successfully!")
    
    # Load prompt audio
    prompt_speech_16k = load_wav('asset/zero_shot_prompt.wav', 16000)
    print("✓ Prompt audio loaded")
    
    # Test different languages
    tests = [
        {
            'lang': 'Portuguese',
            'text': 'Olá! Bem-vindo ao teste de síntese de voz do CosyVoice. Este sistema converte texto em fala de forma natural.',
            'prompt': 'Teste de voz em português.',
            'output': 'output_portuguese.wav'
        },
        {
            'lang': 'English', 
            'text': 'Hello! This is a test of the CosyVoice text-to-speech system. It can generate natural sounding speech.',
            'prompt': 'Testing English speech synthesis.',
            'output': 'output_english.wav'
        },
        {
            'lang': 'Chinese',
            'text': '你好！这是CosyVoice语音合成系统的测试。它可以生成自然的语音。',
            'prompt': '测试中文语音合成。',
            'output': 'output_chinese.wav'
        }
    ]
    
    for test in tests:
        print(f"\n{'='*50}")
        print(f"Testing {test['lang']}:")
        print(f"Text: {test['text']}")
        print(f"Generating speech...")
        
        # Generate speech
        for i, j in enumerate(cosyvoice.inference_zero_shot(
            test['text'], 
            test['prompt'], 
            prompt_speech_16k, 
            stream=False
        )):
            # Save audio
            torchaudio.save(test['output'], j['tts_speech'], cosyvoice.sample_rate)
            
            # Show info
            duration = j['tts_speech'].shape[1] / cosyvoice.sample_rate
            size_kb = os.path.getsize(test['output']) / 1024
            
            print(f"✓ Generated: {test['output']}")
            print(f"  Duration: {duration:.2f} seconds")
            print(f"  Size: {size_kb:.1f} KB")
            print(f"  Sample rate: {cosyvoice.sample_rate} Hz")
    
    # Test streaming mode
    print(f"\n{'='*50}")
    print("Testing streaming mode:")
    text = "Este é um teste do modo streaming, que permite gerar áudio em tempo real com baixa latência."
    print(f"Text: {text}")
    
    chunk_count = 0
    total_duration = 0
    for i, j in enumerate(cosyvoice.inference_zero_shot(text, "Modo streaming.", prompt_speech_16k, stream=True)):
        chunk_count += 1
        chunk_duration = j['tts_speech'].shape[1] / cosyvoice.sample_rate
        total_duration += chunk_duration
        print(f"  Chunk {chunk_count}: {chunk_duration:.3f}s")
        
        # Save first chunk as example
        if chunk_count == 1:
            torchaudio.save('output_streaming_chunk1.wav', j['tts_speech'], cosyvoice.sample_rate)
    
    print(f"✓ Generated {chunk_count} chunks, total duration: {total_duration:.2f}s")
    
    print(f"\n{'='*50}")
    print("✅ All tests completed successfully!")
    
    # Summary
    print("\n📁 Generated files:")
    for f in sorted(os.listdir('.')):
        if f.startswith('output_') and f.endswith('.wav'):
            size = os.path.getsize(f) / 1024
            print(f"  - {f} ({size:.1f} KB)")
    
    print("\n💡 To download files to your local machine:")
    print("scp -P 40053 -i ~/.ssh/id_ed25519 [email protected]:~/CosyVoice/output_*.wav .")
    
except Exception as e:
    print(f"\n❌ Error: {e}")
    import traceback
    traceback.print_exc()