添加一个参考音频筛选工具

GPT_SoVITS/text/symbols.py

@@ -398,4 +398,5 @@
 symbols = [pad] + c + v + ja_symbols + pu_symbols + list(arpa)
 symbols = sorted(set(symbols))
 if __name__ == "__main__":
+    print(symbols)
     print(len(symbols))

Ref_Audio_Selector/common/common.py

@@ -0,0 +1,156 @@
+from tools import my_utils
+from config import python_exec, is_half
+import subprocess
+import sys
+import os
+
+
+class RefAudioListManager:
+    def __init__(self, root_dir):
+        self.audio_dict = {'default': []}
+        absolute_root = os.path.abspath(root_dir)
+
+        for subdir, dirs, files in os.walk(absolute_root):
+            relative_path = os.path.relpath(subdir, absolute_root)
+
+            if relative_path == '.':
+                category = 'default'
+            else:
+                category = relative_path.replace(os.sep, '')
+
+            for file in files:
+                if file.endswith('.wav'):
+                    # 将相对路径转换为绝对路径
+                    audio_abs_path = os.path.join(subdir, file)
+                    if category not in self.audio_dict:
+                        self.audio_dict[category] = []
+                    self.audio_dict[category].append(audio_abs_path)
+
+    def get_audio_list(self):
+        return self.audio_dict
+
+    def get_flattened_audio_list(self):
+        all_audio_files = []
+        for category_audios in self.audio_dict.values():
+            all_audio_files.extend(category_audios)
+        return all_audio_files
+
+    def get_ref_audio_list(self):
+        audio_info_list = []
+        for category, audio_paths in self.audio_dict.items():
+            for audio_path in audio_paths:
+                filename_without_extension = os.path.splitext(os.path.basename(audio_path))[0]
+                audio_info = {
+                    'emotion': f"{category}-{filename_without_extension}",
+                    'ref_path': audio_path,
+                    'ref_text': filename_without_extension,
+                }
+                audio_info_list.append(audio_info)
+        return audio_info_list
+
+
+def batch_clean_paths(paths):
+    """
+    批量处理路径列表，对每个路径调用 clean_path() 函数。
+
+    参数:
+        paths (list[str]): 包含待处理路径的列表。
+
+    返回:
+        list[str]: 经过 clean_path() 处理后的路径列表。
+    """
+    cleaned_paths = []
+    for path in paths:
+        cleaned_paths.append(my_utils.clean_path(path))
+    return cleaned_paths
+
+
+def read_text_file_to_list(file_path):
+    # 按照UTF-8编码打开文件（确保能够正确读取中文）
+    with open(file_path, mode='r', encoding='utf-8') as file:
+        # 读取所有行并存储到一个列表中
+        lines = file.read().splitlines()
+    return lines
+
+
+def get_filename_without_extension(file_path):
+    """
+    Given a file path string, returns the file name without its extension.
+
+    Parameters:
+    file_path (str): The full path to the file.
+
+    Returns:
+    str: The file name without its extension.
+    """
+    base_name = os.path.basename(file_path)  # Get the base name (file name with extension)
+    file_name, file_extension = os.path.splitext(base_name)  # Split the base name into file name and extension
+    return file_name  # Return the file name without extension
+
+
+def read_file(file_path):
+    # 使用with语句打开并读取文件
+    with open(file_path, 'r', encoding='utf-8') as file:  # 'r' 表示以读取模式打开文件
+        # 一次性读取文件所有内容
+        file_content = file.read()
+
+    # 文件在with语句结束时会自动关闭
+    # 现在file_content变量中存储了文件的所有文本内容
+    return file_content
+
+
+def write_text_to_file(text, output_file_path):
+    try:
+        with open(output_file_path, 'w', encoding='utf-8') as file:
+            file.write(text)
+    except IOError as e:
+        print(f"Error occurred while writing to the file: {e}")
+    else:
+        print(f"Text successfully written to file: {output_file_path}")
+
+
+def check_path_existence_and_return(path):
+    """
+    检查给定路径（文件或目录）是否存在。如果存在，返回该路径；否则，返回空字符串。
+    :param path: 待检查的文件或目录路径（字符串）
+    :return: 如果路径存在，返回原路径；否则，返回空字符串
+    """
+    if os.path.exists(path):
+        return path
+    else:
+        return ""
+
+
+def open_file(filepath):
+    if sys.platform.startswith('darwin'):
+        subprocess.run(['open', filepath])  # macOS
+    elif os.name == 'nt':  # For Windows
+        os.startfile(filepath)
+    elif os.name == 'posix':  # For Linux, Unix, etc.
+        subprocess.run(['xdg-open', filepath])
+
+
+def start_new_service(script_path):
+    # 对于Windows系统
+    if sys.platform.startswith('win'):
+        cmd = f'start cmd /k {python_exec} {script_path}'
+    # 对于Mac或者Linux系统
+    else:
+        cmd = f'xterm -e {python_exec} {script_path}'
+
+    proc = subprocess.Popen(cmd, shell=True)
+
+    # 关闭之前启动的子进程
+    # proc.terminate()
+
+    # 或者如果需要强制关闭可以使用
+    # proc.kill()
+
+    return proc
+
+
+if __name__ == '__main__':
+    dir = r'C:\Users\Administrator\Desktop/test'
+    dir2 = r'"C:\Users\Administrator\Desktop\test2"'
+    dir, dir2 = batch_clean_paths([dir, dir2])
+    print(dir, dir2)

Ref_Audio_Selector/common/model_manager.py

@@ -0,0 +1,46 @@
+import os
+import re
+
+pretrained_sovits_name = "GPT_SoVITS/pretrained_models/s2G488k.pth"
+pretrained_gpt_name = "GPT_SoVITS/pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt"
+SoVITS_weight_root = "SoVITS_weights"
+GPT_weight_root = "GPT_weights"
+os.makedirs(SoVITS_weight_root, exist_ok=True)
+os.makedirs(GPT_weight_root, exist_ok=True)
+
+speaker_verification_models = {
+    'speech_campplus_sv_zh-cn_16k-common': {
+        'task': 'speaker-verification',
+        'model': 'Ref_Audio_Selector/tool/speaker_verification/models/speech_campplus_sv_zh-cn_16k-common',
+        'model_revision': 'v1.0.0'
+    },
+    'speech_eres2net_sv_zh-cn_16k-common': {
+        'task': 'speaker-verification',
+        'model': 'Ref_Audio_Selector/tool/speaker_verification/models/speech_eres2net_sv_zh-cn_16k-common',
+        'model_revision': 'v1.0.5'
+    }
+}
+
+def custom_sort_key(s):
+    # 使用正则表达式提取字符串中的数字部分和非数字部分
+    parts = re.split('(\d+)', s)
+    # 将数字部分转换为整数，非数字部分保持不变
+    parts = [int(part) if part.isdigit() else part for part in parts]
+    return parts
+
+
+def get_gpt_model_names():
+    gpt_names = [pretrained_gpt_name]
+    for name in os.listdir(GPT_weight_root):
+        if name.endswith(".ckpt"): gpt_names.append("%s/%s" % (GPT_weight_root, name))
+    sorted(gpt_names, key=custom_sort_key)
+    return gpt_names
+
+
+def get_sovits_model_names():
+    sovits_names = [pretrained_sovits_name]
+    for name in os.listdir(SoVITS_weight_root):
+        if name.endswith(".pth"): sovits_names.append("%s/%s" % (SoVITS_weight_root, name))
+    sorted(sovits_names, key=custom_sort_key)
+    return sovits_names
+

Ref_Audio_Selector/common/time_util.py

@@ -0,0 +1,72 @@
+import time
+import os
+from Ref_Audio_Selector.config_param.log_config import p_logger
+import Ref_Audio_Selector.config_param.config_params as params
+
+
+def timeit_decorator(func):
+    """
+    装饰器，用于计算被装饰函数的执行时间。
+
+    参数:
+        func (function): 要计时的函数。
+
+    返回:
+        function: 包含计时功能的新函数。
+    """
+
+    def wrapper(*args, **kwargs):
+        if params.time_log_print_type != 'file':
+            return func(*args, **kwargs)
+
+        start_time = time.perf_counter()  # 使用 perf_counter 获取高精度计时起点
+
+        func_result = func(*args, **kwargs)  # 执行原函数
+
+        end_time = time.perf_counter()  # 获取计时终点
+        elapsed_time = end_time - start_time  # 计算执行耗时
+
+        # 记录日志内容
+        log_message = f"进程ID: {os.getpid()}, {func.__name__} 执行耗时: {elapsed_time:.6f} 秒"
+        p_logger.info(log_message)
+
+        return func_result
+
+    return wrapper
+
+
+def time_monitor(func):
+    """
+        返回结果，追加时间
+    """
+
+    def wrapper(*args, **kwargs):
+
+        start_time = time.perf_counter()  # 使用 perf_counter 获取高精度计时起点
+
+        func_result = func(*args, **kwargs)  # 执行原函数
+
+        end_time = time.perf_counter()  # 获取计时终点
+        elapsed_time = end_time - start_time  # 计算执行耗时
+
+        return elapsed_time, func_result
+
+    return wrapper
+
+
+# 使用装饰器
+@timeit_decorator
+def example_function(n):
+    time.sleep(n)  # 假设这是需要计时的函数，这里模拟耗时操作
+    return n * 2
+
+
+def example_function2(n):
+    time.sleep(n)  # 假设这是需要计时的函数，这里模拟耗时操作
+    return n * 2
+
+
+if __name__ == "__main__":
+    # 调用经过装饰的函数
+    # result = example_function(2)
+    print(time_monitor(example_function2)(2))

Ref_Audio_Selector/config.ini

@@ -0,0 +1,57 @@
+# config.ini
+
+[Base]
+# 服务端口号
+server_port = 9423
+# 参考音频目录
+reference_audio_dir = refer_audio
+# 临时文件目录
+temp_dir = Ref_Audio_Selector/temp
+
+[Log]
+# 日志保存目录路径
+log_dir = Ref_Audio_Selector/log/general
+# 日志级别 CRITICAL、FATAL、ERROR、WARNING、WARN、INFO、DEBUG、NOTSET、
+log_level = INFO
+# 函数时间消耗日志打印类型 file 打印到文件; close 关闭
+time_log_print_type = file
+# 函数时间消耗日志保存目录路径
+time_log_print_dir = Ref_Audio_Selector/log/performance
+
+[AudioSample]
+# list转换待选参考音频目录
+list_to_convert_reference_audio_dir = refer_audio_all
+# 音频相似度目录
+audio_similarity_dir = similarity
+# 是否开启基准音频预采样 true false
+enable_pre_sample = true
+
+[Inference]
+# 默认测试文本位置
+default_test_text_path = Ref_Audio_Selector/file/test_content/test_content.txt
+# 推理音频目录
+inference_audio_dir = inference_audio
+# 推理音频文本聚合目录
+inference_audio_text_aggregation_dir = text
+# 推理音频情绪聚合目录
+inference_audio_emotion_aggregation_dir = emotion
+
+[ResultCheck]
+# asr输出文件
+asr_filename = asr
+# 文本相似度输出目录
+text_similarity_output_dir = text_similarity
+# 文本情绪平均相似度报告文件名
+text_emotion_average_similarity_report_filename = average_similarity
+# 文本相似度按情绪聚合明细文件名
+text_similarity_by_emotion_detail_filename = emotion_group_detail
+# 文本相似度按文本聚合明细文件名
+text_similarity_by_text_detail_filename = text_group_detail
+
+[AudioConfig]
+# 默认模板文件位置
+default_template_path = Ref_Audio_Selector/file/config_template/ref_audio_template.txt
+# 参考音频配置文件名
+reference_audio_config_filename = refer_audio
+
+[Other]