ChatTTS app.py Demo

gistfile1.txt

import os
import numpy as np
import platform
import subprocess
import wave
import torch
from ChatTTS import Chat
import argparse


def save_audio_to_wav(audio_data, file_path, sample_rate=24000):
    # 将音频数据转换为16位PCM格式
    audio_data = (audio_data / np.max(np.abs(audio_data)) * 32767).astype(np.int16)
    with wave.open(file_path, "wb") as wf:
        wf.setnchannels(1)  # 单通道
        wf.setsampwidth(2)  # 16位
        wf.setframerate(sample_rate)
        wf.writeframes(audio_data.tobytes())


def open_folder(path):
    """
    Open the folder at the specified path in the default file explorer.
    """
    if platform.system() == "Windows":
        # For Windows, use 'explorer'.
        subprocess.run(['explorer', path], check=True)
    elif platform.system() == "Darwin":
        # For macOS, use 'open' to open Finder at the given path.
        subprocess.run(['open', path], check=True)
    else:
        # Optionally, handle Linux systems with 'xdg-open'
        subprocess.run(['xdg-open', path], check=True)


# Setup argument parser
parser = argparse.ArgumentParser(description="Generate TTS with specific speaker settings.")
parser.add_argument('--rand', action='store_true', help='Randomly generate speaker embedding.')
args = parser.parse_args()

# Initialize and load the TTS model
chat = Chat()
chat.load_models()

# Ensure output directory exists
output_dir = "./output"
os.makedirs(output_dir, exist_ok=True)

# Determine speaker embedding
speaker_embedding_path = './output/speaker_embedding.pt'
if args.rand or not os.path.exists(speaker_embedding_path):
    # Generate random speaker
    std, mean = torch.load('./ChatTTS/assets/spk_stat.pt').chunk(2)
    rand_spk = torch.randn(768) * std + mean
    torch.save(rand_spk, speaker_embedding_path)
    print("Generated and saved new speaker embedding.")
else:
    # Load existing speaker
    rand_spk = torch.load(speaker_embedding_path)
    print("Loaded existing speaker embedding.")

# Parameters for inference
params_infer_code = {
    'spk_emb': rand_spk,
    'temperature': 0.5,
    'top_P': 0.8,
    'top_K': 40
}

params_refine_text = {
    'prompt': '[oral_2][laugh_0][break_4]'  # 更自然的口语化，适度的笑声和较长的停顿
}

# Adjusted text content, combined with elements of Children's Day
texts = """
小朋友们,
儿童节快乐。[uv_break]
今天你们有没有收到很多礼物呢,
让我们永远保持童真，[uv_break]
永远年轻，永远夜里尿炕[laugh]。
""".replace('\n', '')

# Use the model for inference
audio_output = chat.infer(texts, params_infer_code=params_infer_code)

# Save the file
output_path = os.path.join(output_dir, "out.wav")
save_audio_to_wav(audio_output, output_path)
print(f"音频文件已保存到：{output_path}")

# Open the output directory
open_folder(output_dir)