test_tts_cosyvoice.py

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

print("=== CosyVoice Text-to-Speech Test ===\n")

# Add Matcha-TTS to path
sys.path.append('third_party/Matcha-TTS')

# Wait for model files
required_files = [
    'pretrained_models/CosyVoice-300M/flow.pt',
    'pretrained_models/CosyVoice-300M/speech_tokenizer_v1.onnx',
    'pretrained_models/CosyVoice-300M/campplus.onnx'
]

print("Checking for required model files...")
max_wait = 300  # 5 minutes
start_time = time.time()

while True:
    missing_files = [f for f in required_files if not os.path.exists(f)]
    if not missing_files:
        print("All required files found!")
        break
    
    elapsed = time.time() - start_time
    if elapsed > max_wait:
        print(f"Timeout waiting for files. Missing: {missing_files}")
        sys.exit(1)
    
    print(f"Waiting for downloads to complete... ({int(elapsed)}s elapsed)")
    time.sleep(10)

try:
    from cosyvoice.cli.cosyvoice import CosyVoice
    from cosyvoice.utils.file_utils import load_wav
    import torchaudio
    
    print("\n1. Testing Zero-Shot Voice Cloning")
    print("-" * 40)
    
    # Initialize model
    cosyvoice = CosyVoice('pretrained_models/CosyVoice-300M', load_jit=False, load_trt=False, fp16=False)
    
    # Test 1: Zero-shot with provided prompt audio
    if os.path.exists('asset/zero_shot_prompt.wav'):
        prompt_speech_16k = load_wav('asset/zero_shot_prompt.wav', 16000)
        
        text_pt = "Olá, este é um teste de síntese de voz em português brasileiro."
        prompt_pt = "Testando a conversão de texto para fala."
        
        print(f"Generating Portuguese speech: '{text_pt}'")
        for i, j in enumerate(cosyvoice.inference_zero_shot(text_pt, prompt_pt, prompt_speech_16k, stream=False)):
            output_file = f'output_portuguese_{i}.wav'
            torchaudio.save(output_file, j['tts_speech'], cosyvoice.sample_rate)
            print(f"✓ Saved: {output_file}")
    
    # Test 2: Chinese speech
    text_zh = "你好，我是通义生成式语音大模型，很高兴为您服务。"
    prompt_zh = "希望这个测试能够成功。"
    
    print(f"\nGenerating Chinese speech: '{text_zh}'")
    if os.path.exists('asset/zero_shot_prompt.wav'):
        for i, j in enumerate(cosyvoice.inference_zero_shot(text_zh, prompt_zh, prompt_speech_16k, stream=False)):
            output_file = f'output_chinese_{i}.wav'
            torchaudio.save(output_file, j['tts_speech'], cosyvoice.sample_rate)
            print(f"✓ Saved: {output_file}")
    
    # Test 3: English speech  
    text_en = "Hello, this is a test of the CosyVoice text-to-speech system. It supports multiple languages."
    prompt_en = "Testing voice synthesis."
    
    print(f"\nGenerating English speech: '{text_en}'")
    if os.path.exists('asset/zero_shot_prompt.wav'):
        for i, j in enumerate(cosyvoice.inference_zero_shot(text_en, prompt_en, prompt_speech_16k, stream=False)):
            output_file = f'output_english_{i}.wav'
            torchaudio.save(output_file, j['tts_speech'], cosyvoice.sample_rate)
            print(f"✓ Saved: {output_file}")
    
    print("\n2. Testing SFT Mode (if available)")
    print("-" * 40)
    
    # Try SFT model if available
    if os.path.exists('pretrained_models/CosyVoice-300M-SFT/cosyvoice.yaml'):
        try:
            cosyvoice_sft = CosyVoice('pretrained_models/CosyVoice-300M-SFT', load_jit=False, load_trt=False, fp16=False)
            
            # List available speakers
            speakers = cosyvoice_sft.list_available_spks()
            print(f"Available speakers: {speakers[:5]}...")  # Show first 5
            
            # Generate with first available speaker
            if speakers:
                text = "Este é um teste usando o modelo SFT com falantes pré-definidos."
                speaker = speakers[0]
                print(f"\nGenerating with speaker '{speaker}': '{text}'")
                
                for i, j in enumerate(cosyvoice_sft.inference_sft(text, speaker, stream=False)):
                    output_file = f'output_sft_{i}.wav'
                    torchaudio.save(output_file, j['tts_speech'], cosyvoice_sft.sample_rate)
                    print(f"✓ Saved: {output_file}")
        except Exception as e:
            print(f"SFT model not ready yet: {e}")
    
    print("\n=== Test completed successfully! ===")
    print("\nGenerated audio files:")
    for f in os.listdir('.'):
        if f.startswith('output_') and f.endswith('.wav'):
            size = os.path.getsize(f) / 1024
            print(f"  - {f} ({size:.1f} KB)")
    
except Exception as e:
    print(f"\nError during test: {e}")
    import traceback
    traceback.print_exc()