__version__ = "2.4.3 240414" import os, json import torch import logging logging.getLogger("markdown_it").setLevel(logging.ERROR) logging.getLogger("urllib3").setLevel(logging.ERROR) logging.getLogger("httpcore").setLevel(logging.ERROR) logging.getLogger("httpx").setLevel(logging.ERROR) logging.getLogger("asyncio").setLevel(logging.ERROR) logging.getLogger("charset_normalizer").setLevel(logging.ERROR) logging.getLogger("torchaudio._extension").setLevel(logging.ERROR) class Inference_Config(): def __init__(self): self.config_path = os.path.join(os.path.dirname(os.path.dirname(__file__)), "config.json") assert os.path.exists(self.config_path), f"配置文件不存在: {self.config_path}" if os.path.exists(self.config_path): with open(self.config_path, 'r', encoding='utf-8') as f: config:dict = json.load(f) self.workers = config.get("workers", 10) self.models_path = config.get("models_path", "trained") self.tts_host = config.get("tts_host", "0.0.0.0") self.tts_port = config.get("tts_port", 5000) self.default_batch_size = config.get("batch_size", 1) self.default_word_count = config.get("max_word_count", 50) self.enable_auth = config.get("enable_auth", "false").lower() == "true" self.is_classic = config.get("classic_inference", "false").lower() == "true" self.is_share = config.get("is_share", "false").lower() == "true" self.max_text_length = config.get("max_text_length", -1) self.disabled_features = config.get("disabled_features", []) self.allowed_adapters = config.get("allowed_adapters", ["gsv_fast", "gsv_classic", "azure"]) self.save_model_cache = config.get("save_model_cache", "false").lower() == "true" self.save_prompt_cache = config.get("save_prompt_cache", "false").lower() == "true" locale_language = str(config.get("locale", "auto")) self.locale_language = None if locale_language.lower() == "auto" else locale_language if self.enable_auth: self.users = config.get("user", {}) self.synthesizer = config.get("synthesizer", "gsv_fast") global inference_config inference_config = Inference_Config() models_path = inference_config.models_path def load_infer_config(character_path): config_path = os.path.join(character_path, "infer_config.json") """加载环境配置文件""" with open(config_path, 'r', encoding='utf-8') as f: json_content = f.read().replace("\\", "/") config = json.loads(json_content) return config def auto_generate_infer_config(character_path): ## TODO: Auto-generate wav-list and prompt-list from character_path ## # Initialize variables for file detection print(f"正在自动生成配置文件: {character_path}") ckpt_file_found = None pth_file_found = None wav_file_found = None # Iterate through files in character_path to find matching file types for dirpath, dirnames, filenames in os.walk(character_path): for file in filenames: # 构建文件的完整路径 full_path = os.path.join(dirpath, file) # 从full_path中移除character_path部分 relative_path = remove_character_path(full_path,character_path) # 根据文件扩展名和变量是否已赋值来更新变量 if file.lower().endswith(".ckpt") and ckpt_file_found is None: ckpt_file_found = relative_path elif file.lower().endswith(".pth") and pth_file_found is None: pth_file_found = relative_path elif file.lower().endswith(".wav") and wav_file_found is None: wav_file_found = relative_path elif file.lower().endswith(".mp3"): import pydub # Convert mp3 to wav wav_file_path = os.path.join(dirpath,os.path.splitext(file)[0] + ".wav") pydub.AudioSegment.from_mp3(full_path).export(wav_file_path, format="wav") if wav_file_found is None: wav_file_found = remove_character_path(os.path.join(dirpath,os.path.splitext(file)[0] + ".wav"),character_path) # Initialize infer_config with gpt_path and sovits_path regardless of wav_file_found infer_config = { "gpt_path": ckpt_file_found, "sovits_path": pth_file_found, "software_version": "1.1", r"简介": r"这是一个配置文件适用于https://github.com/X-T-E-R/TTS-for-GPT-soVITS，是一个简单好用的前后端项目" } # If wav file is also found, update infer_config to include ref_wav_path, prompt_text, and prompt_language if wav_file_found: wav_file_name = os.path.splitext(os.path.basename(wav_file_found))[0] # Extract the filename without extension infer_config["emotion_list"] = { "default": { "ref_wav_path": wav_file_found, "prompt_text": wav_file_name, "prompt_language": "多语种混合" } } else: raise Exception("找不到wav参考文件！请把有效wav文件放置在模型文件夹下。") pass # Check if the essential model files were found if ckpt_file_found and pth_file_found: infer_config_path = os.path.join(character_path, "infer_config.json") try: with open(infer_config_path , 'w', encoding='utf-8') as f: json.dump(infer_config, f, ensure_ascii=False, indent=4) except IOError as e: print(f"无法写入文件: {infer_config_path}. 错误: {e}") return infer_config_path else: return "Required model files (.ckpt or .pth) not found in character_path directory." def update_character_info(models_path:str=None): # try: # with open(os.path.join(models_path, "character_info.json"), "r", encoding='utf-8') as f: # default_character = json.load(f).get("deflaut_character", None) # except: if models_path in [None, ""]: models_path = inference_config.models_path default_character = "" characters_and_emotions = {} for character_subdir in [f for f in os.listdir(models_path) if os.path.isdir(os.path.join(models_path, f))]: character_subdir = character_subdir if os.path.exists(os.path.join(models_path, character_subdir, "infer_config.json")): try: with open(os.path.join(models_path, character_subdir, "infer_config.json"), "r", encoding='utf-8') as f: config = json.load(f) emotion_list=[emotion for emotion in config.get('emotion_list', None)] if emotion_list is not None: characters_and_emotions[character_subdir] = emotion_list else: characters_and_emotions[character_subdir] = ["default"] except: characters_and_emotions[character_subdir] = ["default"] else: characters_and_emotions[character_subdir] = ["default"] return {"deflaut_character": default_character, "characters_and_emotions": characters_and_emotions} def test_fp16_computation(): # 检查CUDA是否可用 if not torch.cuda.is_available(): return False, "CUDA is not available. Please check your installation." try: # 创建一个简单的半精度张量计算任务 # 例如，执行一个半精度的矩阵乘法 a = torch.randn(3, 3, dtype=torch.float16).cuda() # 将张量a转换为半精度并移动到GPU b = torch.randn(3, 3, dtype=torch.float16).cuda() # 将张量b转换为半精度并移动到GPU c = torch.matmul(a, b) # 执行半精度的矩阵乘法 # 如果没有发生错误，我们认为GPU支持半精度运算 return True, "Your GPU supports FP16 computation." except Exception as e: # 如果执行过程中发生异常，我们认为GPU不支持半精度运算 return False, f"Your GPU does not support FP16 computation. Error: {e}" def get_device_info(): global device, is_half try: return device, is_half except: if torch.cuda.is_available(): device = "cuda" is_half = True else: device = "cpu" is_half = False # 取得模型文件夹路径 config_path = os.path.join(os.path.dirname(os.path.dirname(__file__)), "config.json") if os.path.exists(config_path): with open(config_path, 'r', encoding='utf-8') as f: _config = json.load(f) if _config.get("device", "auto") != "auto": device = _config["device"] is_half = (device == "cpu") if _config.get("half_precision", "auto") != "auto": is_half = _config["half_precision"].lower() == "true" supports_fp16, message = test_fp16_computation() if not supports_fp16 and is_half: is_half = False print(message) return device, is_half def remove_character_path(full_path,character_path): # 从full_path中移除character_path部分 return os.path.relpath(full_path, character_path)