fix ASR

docs/cn/README.md

@@ -204,6 +204,9 @@ python tools/uvr5/webui.py "<infer_device>" <is_half> <webui_port_uvr5>
 ````
 python mdxnet.py --model --input_root --output_vocal --output_ins --agg_level --format --device --is_half_precision 
 ````
+
+---
+
 这是使用命令行完成数据集的音频切分的方式
 ````
 python audio_slicer.py \
@@ -214,17 +217,38 @@ python audio_slicer.py \
     --min_interval <shortest_time_gap_between_adjacent_subclips> 
     --hop_size <step_size_for_computing_volume_curve>
 ````
-这是使用命令行完成数据集ASR处理的方式（仅限中文）
-````
-python tools/damo_asr/cmd-asr.py "<Path to the directory containing input audio files>"
-````
-通过Faster_Whisper进行ASR处理（除中文之外的ASR标记）
 
-（没有进度条，GPU性能可能会导致时间延迟）
+---
+
+这是使用命令行完成数据集ASR处理的方式
+
+通过 FunASR 进行 ASR 处理 (仅中文)
 ````
-python ./tools/damo_asr/WhisperASR.py -i <input> -o <output> -f <file_name.list> -l <language>
+python tools/asr/funasr_asr.py \
+  -i "<Path of single audio file or Path to the directory containing input audio files>" \
+  -o "Output folder"
 ````
-启用自定义列表保存路径
+
+参数: 
+- `-i/--input_file_or_folder`: 输入音频文件/含有音频的文件夹路径.
+- `-o/--output_folder`: 输出文件夹路径, 输出文件将以输入文件所在目录名称命名, 后缀为 `.list`.
+- ~~`-s/--model_size`~~: 模型尺寸, 基于 Paraformer-Large, 其他输入目前无效.
+- ~~`-l/--language`~~: 识别语言, 仅支持中文, 其他输入目前无效.
+- ~~`-p/--precision`~~: 计算精度, 其他输入目前无效.
+
+通过 Faster_Whisper 进行 ASR 处理 (其他语言, 中文会自动转到 FunASR 进行处理)
+
+````
+python tools/asr/fasterwhisper_asr.py -i <input_file_or_folder> -o <output_folder> -s <model_size> -l <language> -p <precision>
+````
+参数:
+- `-i/--input_file_or_folder`: 输入音频文件/含有音频的文件夹路径.
+- `-o/--output_folder`: 输出文件夹路径, 输出文件将以输入文件所在目录名称命名, 后缀为 `.list`.
+- `-s/--model_size`: 模型尺寸, 可选值与本地是否存在模型相关, 默认为 `large-v3`.
+- `-l/--language`: 识别语言, 默认为 `auto` 自动识别语言.
+- `-p/--precision`: 计算精度, 可选 `fp16`, `fp32`, 使用 CPU 时会自动调整.
+
+
 ## 致谢
 
 特别感谢以下项目和贡献者：

tools/asr/config.py

@@ -1,31 +1,114 @@
 import os
+from datetime import datetime
 
-def check_fw_local_models():
-    '''
-    启动时检查本地是否有 Faster Whisper 模型.
-    '''
-    model_size_list = [
-        "tiny",     "tiny.en", 
-        "base",     "base.en", 
-        "small",    "small.en", 
-        "medium",   "medium.en", 
-        "large",    "large-v1", 
-        "large-v2", "large-v3"]
-    for i, size in enumerate(model_size_list):
-        if os.path.exists(f'tools/asr/models/faster-whisper-{size}'):
-            model_size_list[i] = size + '-local'
-    return model_size_list
+from tqdm import tqdm
+from tqdm.contrib.logging import logging_redirect_tqdm
+
+from tools.my_utils import ASR_Logger
+
+class BaseASR:
+    def __init__(self):
+        pass
+
+    def check_local_model(self, model_name, model_file, cache_path):
+        '''
+        启动时检查本地是否有模型文件夹.
+        '''
+        # 先检查当前项目是否有模型文件夹
+        local_path = os.path.normpath('tools/asr/models')
+        model_path = ''
+        flag = ''
+        for root, dirs, files in os.walk(local_path):
+            if model_file in files and model_name + os.sep in os.path.join(root, model_file):
+                model_path, flag = root, 'local'
+        if not model_path:
+            # 当前项目没有则检索本地缓存
+            for root, dirs, files in os.walk(cache_path):
+                if model_file in files and model_name + os.sep in os.path.join(root, model_file):
+                    model_path, flag = root, 'cache'
+        return model_path, flag
+        
+    def load_model(self):
+        """
+        加载模型.
+        """
+        raise NotImplementedError
+    
+    def inference(self):
+        """
+        对单个文件进行推理, 返回文本, 和相应的语言.
+        """
+        raise NotImplementedError
+
+    def inference_file_or_folder(self, input_file_or_folder, output_folder, language):
+        """
+        对文件夹/文件进行推理, 并保存结果.
+        """
+        assert os.path.exists(input_file_or_folder), ASR_Logger.error('输入路径不存在.')
+        if os.path.isfile(input_file_or_folder):
+            # 若为文件获取其父目录的文件名
+            file_path = input_file_or_folder
+            input_file_paths = [os.path.abspath(file_path)]
+            output_file_name = os.path.basename(os.path.dirname(file_path))
+        else:
+            input_folder = input_file_or_folder
+            input_file_names = os.listdir(input_folder)
+            input_file_names.sort()
+            input_file_paths = []
+            for input_file_name in input_file_names:
+                input_file_path = os.path.abspath(os.path.join(input_folder, input_file_name))
+                if os.path.isfile(input_file_path):
+                    input_file_paths.append(input_file_path) 
+
+            output_file_name = os.path.basename(input_folder)
+
+        result = []
+
+        if not os.path.exists(output_folder):
+            os.makedirs(output_folder)
+        
+        if language == 'auto':
+            language = None #不设置语种由模型自动输出概率最高的语种
+        ASR_Logger.info("开始转写")
+        with logging_redirect_tqdm([ASR_Logger]):
+            for file_path in tqdm(input_file_paths, desc="转写进度 ", dynamic_ncols=True):
+                text, output_language = self.inference(file_path, language)
+                if text and output_language:
+                    result.append(f"{file_path}|{output_file_name}|{output_language.upper()}|{text}")
+        if not result:
+            ASR_Logger.error("没有转写结果, 放弃保存.")
+            return 
+
+        output_file_path = os.path.abspath(f'{output_folder}/{output_file_name}.list')
+        if os.path.exists(output_file_path):
+            ASR_Logger.info('输出文件路径已存在, 文件名添加时间戳.')
+            timestamp = datetime.now().strftime('%Y%m%d%H%M%S')
+            file_name, file_extension = os.path.splitext(output_file_path)
+            output_file_path = f"{file_name}-{timestamp}{file_extension}"
+        with open(output_file_path, "w", encoding="utf-8") as f:
+            f.write("\n".join(result))
+            ASR_Logger.info(f"任务完成->标注文件路径: {output_file_path}\n")
+        return output_file_path
+    
+fw_model_size_list = [
+    "tiny",     "tiny.en", 
+    "base",     "base.en", 
+    "small",    "small.en", 
+    "medium",   "medium.en", 
+    "large",    "large-v1", 
+    "large-v2", "large-v3"]
 
 asr_dict = {
     "达摩 ASR (中文)": {
+        'name': 'funasr',
         'lang': ['zh'],
         'size': ['large'],
         'path': 'funasr_asr.py',
     },
     "Faster Whisper (多语种)": {
+        'name': 'fasterwhisper',
         'lang': ['auto', 'zh', 'en', 'ja'],
-        'size': check_fw_local_models(),
+        'size': fw_model_size_list,
         'path': 'fasterwhisper_asr.py'
     }
-}
-
+}

tools/asr/fasterwhisper_asr.py

@@ -1,16 +1,15 @@
 import argparse
 import os
-os.environ["HF_ENDPOINT"]="https://hf-mirror.com"
 import traceback
-import requests
-from glob import glob
 
+import torch
 from faster_whisper import WhisperModel
-from tqdm import tqdm
 
-from tools.asr.config import check_fw_local_models
-from tools.asr.funasr_asr import only_asr
+from tools.asr.config import fw_model_size_list, BaseASR
+from tools.asr.funasr_asr import FunASR
+from tools.my_utils import ASR_Logger
 
+os.environ["HF_ENDPOINT"]="https://hf-mirror.com"
 os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
 
 language_code_list = [
@@ -36,72 +35,106 @@ language_code_list = [
     "vi", "yi", "yo", "zh", "yue",
     "auto"]
 
-def execute_asr(input_folder, output_folder, model_size, language,precision):
-    if '-local' in model_size:
-        model_size = model_size[:-6]
-        model_path = f'tools/asr/models/faster-whisper-{model_size}'
-    else:
-        model_path = model_size
-    if language == 'auto':
-        language = None #不设置语种由模型自动输出概率最高的语种
-    print("loading faster whisper model:",model_size,model_path)
-    try:
-        model = WhisperModel(model_path, device="cuda", compute_type=precision)
-    except:
-        return print(traceback.format_exc())
-    output = []
-    output_file_name = os.path.basename(input_folder)
-    output_file_path = os.path.abspath(f'{output_folder}/{output_file_name}.list')
+class FasterWhisperASR(BaseASR):
 
-    if not os.path.exists(output_folder):
-        os.makedirs(output_folder)
+    def __init__(self, model_size, device="cuda", precision="float16"):
+        device, precision = [device, precision] if torch.cuda.is_available() else ["cpu", "float32"]
+        self.check_local_models()
+        self.model = self.load_model(model_size, device, precision)
+        assert self.model is not None, ASR_Logger.error('模型不存在')
+        self.zh_model = None
+
+    @classmethod
+    def check_local_models(self):
+        '''
+        启动时检查本地是否有 Faster Whisper 模型.
+        '''
+        self.model_size_list = fw_model_size_list.copy()
+        self.model_path_dict = {}
+        for i, size in enumerate(self.model_size_list):
+            model_name = f"faster-whisper-{size}"
+            model_path, flag = super().check_local_model(
+                self, 
+                model_name = model_name, 
+                model_file = 'model.bin', 
+                cache_path = os.path.normpath(os.path.expanduser(f"~/.cache/huggingface/hub/")))
+            if flag:
+                self.model_size_list[i] = f"{size}-{flag}"
+                self.model_path_dict[self.model_size_list[i]] = model_path
+        return self.model_size_list
+
+    def load_model(self, model_size, device="cuda", precision="float16"):
+        if '-local' in model_size or '-cache' in model_size:
+            model_path = self.model_path_dict[model_size]
+            model_size = model_size[:-6]
+            ASR_Logger.info(f"加载模型: 从 {model_path} 加载 faster-whisper-{model_size} 模型.")
+            if 'huggingface' in model_path:
+                ASR_Logger.warning(f"可将 {model_path} 移动到 tools/asr/models/ 文件夹下并重命名为 faster-whisper-{model_size}.")
+        else:
+            model_path = model_size
+            ASR_Logger.warning(f"下载模型: 从 https://hf-mirror.com/Systran/faster-whisper-{model_size} 下载 faster-whisper-{model_size} 模型.")
 
-    for file in tqdm(glob(os.path.join(input_folder, '**/*.wav'), recursive=True)):
         try:
-            segments, info = model.transcribe(
-                audio          = file,
+            model = WhisperModel(model_path, device=device, compute_type=precision)
+            if model.model.device != 'cpu':
+                device_name = torch.cuda.get_device_name(model.model.device)
+            else:
+                device_name = 'CPU'
+            ASR_Logger.info(f"运行设备: {device_name}, 设定精度: {precision}.")
+            ASR_Logger.info(f"创建模型: Faster Whisper 完成.\n")
+            return model
+        except:
+            ASR_Logger.info(traceback.format_exc())
+            ASR_Logger.error(f"模型加载失败 or 下载失败, 可访问 https://hf-mirror.com/Systran/faster-whisper-{model_size} 自行下载, 并放置于 tools/asr/models/ 文件夹下")
+            return 
+
+    def inference(self, file_path, language='auto'):
+        try:
+            if language == 'auto': 
+                language = None
+
+            segments, info = self.model.transcribe(
+                audio          = file_path,
                 beam_size      = 5,
                 vad_filter     = True,
                 vad_parameters = dict(min_silence_duration_ms=700),
                 language       = language)
-            text = ''
 
             if info.language == "zh":
-                print("检测为中文文本,转funasr处理")
-                text = only_asr(file)
-
-            if text == '':
-                for segment in segments:
-                    text += segment.text
-            output.append(f"{file}|{output_file_name}|{info.language.upper()}|{text}")
+                ASR_Logger.info("检测为中文文本, 转 FunASR 处理.")
+                if self.zh_model is None:
+                    self.zh_model = FunASR()
+                text, language = self.zh_model.inference(file_path)
+            else:
+                text = ''.join([segment.text for segment in segments])
+            return text, info.language
         except:
-            return print(traceback.format_exc())
-        
-    with open(output_file_path, "w", encoding="utf-8") as f:
-        f.write("\n".join(output))
-        print(f"ASR 任务完成->标注文件路径: {output_file_path}\n")
-    return output_file_path
+            ASR_Logger.error(f"当前文件 {file_path} 转写失败, 可能不是有效的音频文件.")
+            ASR_Logger.error(traceback.format_exc())
+            return '', ''
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
-    parser.add_argument("-i", "--input_folder", type=str, required=True,
-                        help="Path to the folder containing WAV files.")
+    parser.add_argument("-i", "--input_file_or_folder", type=str, required=True,
+                        help="Input audio file path or folder contain audio files.")
     parser.add_argument("-o", "--output_folder", type=str, required=True, 
                         help="Output folder to store transcriptions.")
     parser.add_argument("-s", "--model_size", type=str, default='large-v3', 
-                        choices=check_fw_local_models(),
+                        choices=FasterWhisperASR.check_local_models(),
                         help="Model Size of Faster Whisper")
-    parser.add_argument("-l", "--language", type=str, default='ja',
+    parser.add_argument("-l", "--language", type=str, default='auto',
                         choices=language_code_list,
                         help="Language of the audio files.")
-    parser.add_argument("-p", "--precision", type=str, default='float16', choices=['float16','float32'],
-                        help="fp16 or fp32")
-
+    parser.add_argument("-p", "--precision", type=str, default='float16', 
+                        choices=['float16','float32'], help="fp16 or fp32")
     cmd = parser.parse_args()
-    output_file_path = execute_asr(
-        input_folder  = cmd.input_folder,
-        output_folder = cmd.output_folder,
-        model_size    = cmd.model_size,
-        language      = cmd.language,
-        precision     = cmd.precision,
+    ASR = FasterWhisperASR(
+        model_size = cmd.model_size,
+        precision  = cmd.precision,
     )
+    ASR.inference_file_or_folder(
+        input_file_or_folder = cmd.input_file_or_folder,
+        output_folder        = cmd.output_folder,
+        language             = cmd.language,
+    )
+

tools/asr/funasr_asr.py

@@ -3,74 +3,114 @@
 import argparse
 import os
 import traceback
-from tqdm import tqdm
 
+import torch
 from funasr import AutoModel
 
-path_asr  = 'tools/asr/models/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch'
-path_vad  = 'tools/asr/models/speech_fsmn_vad_zh-cn-16k-common-pytorch'
-path_punc = 'tools/asr/models/punc_ct-transformer_zh-cn-common-vocab272727-pytorch'
-path_asr  = path_asr  if os.path.exists(path_asr)  else "iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch"
-path_vad  = path_vad  if os.path.exists(path_vad)  else "iic/speech_fsmn_vad_zh-cn-16k-common-pytorch"
-path_punc = path_punc if os.path.exists(path_punc) else "iic/punc_ct-transformer_zh-cn-common-vocab272727-pytorch"
+from tools.asr.config import BaseASR
+from tools.my_utils import ASR_Logger
 
-model = AutoModel(
-    model               = path_asr,
-    model_revision      = "v2.0.4",
-    vad_model           = path_vad,
-    vad_model_revision  = "v2.0.4",
-    punc_model          = path_punc,
-    punc_model_revision = "v2.0.4",
-)
+funasr_component = {
+    'asr': {
+        'name': 'Paraformer-Large',
+        'size': 'speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch',
+    },
+    'vad': {
+        'name': 'FSMN-Monophone VAD',
+        'size': 'speech_fsmn_vad_zh-cn-16k-common-pytorch',
+    },
+    'punc': {
+        'name': 'Controllable Time-delay Transformer',
+        'size': 'punc_ct-transformer_zh-cn-common-vocab272727-pytorch',
+    }
+}
 
-def only_asr(input_file):
-    try:
-        text = model.generate(input=input_file)[0]["text"]
-    except:
-        text = ''
-        print(traceback.format_exc())
-    return text
+class FunASR(BaseASR):
+    def __init__(self, model_size='large', device="cuda", precision="float16"):
+        self.check_local_models()
+        self.model = self.load_model()
+        assert self.model is not None, ASR_Logger.error('模型不存在')
 
-def execute_asr(input_folder, output_folder, model_size, language):
-    input_file_names = os.listdir(input_folder)
-    input_file_names.sort()
-    
-    output = []
-    output_file_name = os.path.basename(input_folder)
+    @classmethod
+    def check_local_models(self):
+        '''
+        启动时检查本地是否有 FunASR 相关模型.
+        '''
+        self.model_path_dict = funasr_component
+        for code, dic in self.model_path_dict.items():
+            model_name = dic['size']
+            model_path, flag = super().check_local_model(
+                self,
+                model_name = model_name,
+                model_file = 'model.pt',
+                cache_path = os.path.normpath(os.path.expanduser(f"~/.cache/modelscope/hub/"))
+            )
+            if model_path:
+                self.model_path_dict[code]['path'] = model_path
+            else:
+                # 没有本地路径时, 路径设置为网络链接
+                self.model_path_dict[code]['path'] = 'iic/' + model_name
+        return self.model_path_dict
 
-    for name in tqdm(input_file_names):
+    def load_model(self):
         try:
-            text = model.generate(input="%s/%s"%(input_folder, name))[0]["text"]
-            output.append(f"{input_folder}/{name}|{output_file_name}|{language.upper()}|{text}")
+            for code, dic in self.model_path_dict.items():
+                if os.path.exists(dic['path']):
+                    ASR_Logger.info(f"加载模型: 从 {dic['path']} 加载 {dic['name']} 模型.")
+                    if 'modelscope' in dic['path']:
+                        ASR_Logger.warning(f"可将 {dic['path']} 移动到 tools/asr/models/ 文件夹下.")
+                else:
+                    ASR_Logger.warning(f"下载模型: 从 {dic['path']} 下载 {dic['name']} 模型.")
+            model = AutoModel(
+                model               = self.model_path_dict['asr']['path'],
+                model_revision      = "v2.0.4",
+                vad_model           = self.model_path_dict['vad']['path'],
+                vad_model_revision  = "v2.0.4",
+                punc_model          = self.model_path_dict['punc']['path'],
+                punc_model_revision = "v2.0.4",
+            )
+            ASR_Logger.propagate = False # 避免 FunASR 库导致打印重复日志
+
+            if model.kwargs['device'] != 'cpu':
+                device_name = torch.cuda.get_device_name(model.kwargs['device'])
+            else:
+                device_name = 'CPU'
+            ASR_Logger.info(f"运行设备: {device_name}, 设定精度: --.")
+            ASR_Logger.info(f"创建模型: FunASR 完成.\n")
+            return model
         except:
-            print(traceback.format_exc())
-
-    output_folder = output_folder or "output/asr_opt"
-    os.makedirs(output_folder, exist_ok=True)
-    output_file_path = os.path.abspath(f'{output_folder}/{output_file_name}.list')
-
-    with open(output_file_path, "w", encoding="utf-8") as f:
-        f.write("\n".join(output))
-        print(f"ASR 任务完成->标注文件路径: {output_file_path}\n")
-    return output_file_path
-
+            ASR_Logger.error(traceback.format_exc())
+            raise ValueError(ASR_Logger.error(f"模型加载失败 or 下载失败, 可访问 https://modelscope.cn/organization/iic 自行下载, 并放置于 tools/asr/models/ 文件夹下"))
+        
+    def inference(self, file_path, language='zh'):
+        try:
+            text = self.model.generate(input=file_path)[0]["text"]
+            return text, language
+        except:
+            ASR_Logger.error(f"当前文件 {file_path} 转写失败, 可能不是有效的音频文件.")
+            ASR_Logger.error(traceback.format_exc())
+            return '', ''
+        
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
-    parser.add_argument("-i", "--input_folder", type=str, required=True,
-                        help="Path to the folder containing WAV files.")
+    parser.add_argument("-i", "--input_file_or_folder", type=str, required=True,
+                        help="Input audio file path or folder contain audio files.")
     parser.add_argument("-o", "--output_folder", type=str, required=True, 
                         help="Output folder to store transcriptions.")
     parser.add_argument("-s", "--model_size", type=str, default='large',
                         help="Model Size of FunASR is Large")
     parser.add_argument("-l", "--language", type=str, default='zh', choices=['zh'],
                         help="Language of the audio files.")
-    parser.add_argument("-p", "--precision", type=str, default='float16', choices=['float16','float32'],
-                        help="fp16 or fp32")#还没接入
+    parser.add_argument("-p", "--precision", type=str, default='float16', 
+                        choices=['float16','float32'], help="fp16 or fp32")#还没接入
 
     cmd = parser.parse_args()
-    execute_asr(
-        input_folder  = cmd.input_folder,
-        output_folder = cmd.output_folder,
-        model_size    = cmd.model_size,
-        language      = cmd.language,
+    ASR = FunASR(
+        model_size = cmd.model_size,
+        precision  = cmd.precision,
+    )
+    ASR.inference_file_or_folder(
+        input_file_or_folder = cmd.input_file_or_folder,
+        output_folder        = cmd.output_folder,
+        language             = cmd.language,
     )

tools/my_utils.py

@@ -1,4 +1,8 @@
-import platform,os,traceback
+import logging
+import os
+import platform
+import traceback
+import sys
 import ffmpeg
 import numpy as np
 
@@ -29,3 +33,46 @@ def clean_path(path_str):
     if platform.system() == 'Windows':
         path_str = path_str.replace('/', '\\')
     return path_str.strip(" ").strip('"').strip("\n").strip('"').strip(" ")
+
+COLORS = {
+    'WARNING' : '\033[33m',        # Yellow
+    'ERROR'   : '\033[31m',        # Red
+    'CRITICAL': '\033[35m',
+    'RESET'   : '\033[0m',         # Reset color
+}
+
+class ColoredConsoleHandler(logging.StreamHandler):
+    def emit(self, record):
+        # 获取日志级别对应的颜色
+        color = COLORS.get(record.levelname, '')
+        # 重置颜色
+        reset = COLORS['RESET']
+        # 设置日志消息的颜色
+        self.setFormatter(logging.Formatter(color + '%(asctime)s - %(name)s - %(levelname)s - %(message)s' + reset))
+        # 输出日志消息
+        super().emit(record)
+
+class Tools_Logger():
+    def __init__(self, logger_name, log_level='info', log_path=None):
+        assert type(log_level) == str and log_level.upper() in ['CRITICAL', 'ERROR', 'WARNING', 'INFO', 'DEBUG']
+        log_level = log_level.upper()
+        self.logger = logging.getLogger(logger_name)
+        self.logger.setLevel(log_level)
+
+        if not self.logger.hasHandlers():
+            ch = ColoredConsoleHandler()
+            ch.setLevel(log_level)
+            # formatter = logging.Formatter()
+            # ch.setFormatter(formatter)
+            self.logger.addHandler(ch)
+
+            # if log_path is not None:
+            #     fh = logging.FileHandler(log_path)
+            #     fh.setLevel(log_level)
+            #     fh.setFormatter(formatter)
+            #     self.logger.addHandler(fh)
+
+    def getLogger(self):
+        return self.logger
+
+ASR_Logger = Tools_Logger('ASR').getLogger()

webui.py

@@ -660,6 +660,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             with gr.Row():
                 if_uvr5 = gr.Checkbox(label=i18n("是否开启UVR5-WebUI"),show_label=True)
                 uvr5_info = gr.Textbox(label=i18n("UVR5进程输出信息"))
+            
             gr.Markdown(value=i18n("0b-语音切分工具"))
             with gr.Row():
                 with gr.Row():
@@ -677,16 +678,17 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
                     alpha=gr.Slider(minimum=0,maximum=1,step=0.05,label=i18n("alpha_mix:混多少比例归一化后音频进来"),value=0.25,interactive=True)
                     n_process=gr.Slider(minimum=1,maximum=n_cpu,step=1,label=i18n("切割使用的进程数"),value=4,interactive=True)
                     slicer_info = gr.Textbox(label=i18n("语音切割进程输出信息"))
-            gr.Markdown(value=i18n("0c-中文批量离线ASR工具"))
+            
+            gr.Markdown(value=i18n("0c-离线批量语音识别工具"))
             with gr.Row():
                 open_asr_button = gr.Button(i18n("开启离线批量ASR"), variant="primary",visible=True)
                 close_asr_button = gr.Button(i18n("终止ASR进程"), variant="primary",visible=False)
-                with gr.Column():
+                with gr.Column(scale=3):
                     with gr.Row():
                         asr_inp_dir = gr.Textbox(
-                            label=i18n("输入文件夹路径"),
-                            value="D:\\GPT-SoVITS\\raw\\xxx",
-                            interactive=True,
+                            label       = i18n("输入音频文件/文件夹路径"),
+                            value       = "D:\\GPT-SoVITS\\raw\\xxx",
+                            interactive = True,
                         )
                         asr_opt_dir = gr.Textbox(
                             label       = i18n("输出文件夹路径"),
@@ -695,35 +697,42 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
                         )
                     with gr.Row():
                         asr_model = gr.Dropdown(
-                            label       = i18n("ASR 模型"),
+                            label       = i18n("ASR 模型选择"),
                             choices     = list(asr_dict.keys()),
                             interactive = True,
-                            value="达摩 ASR (中文)"
+                            value       = "达摩 ASR (中文)",
+                            info        = '右侧两项跟随此项更新'
                         )
                         asr_size = gr.Dropdown(
                             label       = i18n("ASR 模型尺寸"),
                             choices     = ["large"],
                             interactive = True,
-                            value="large"
+                            value       = "large",
+                            info        = 'local (当前项目)/cache (本地缓存)/自动下载'
                         )
                         asr_lang = gr.Dropdown(
                             label       = i18n("ASR 语言设置"),
                             choices     = ["zh"],
                             interactive = True,
-                            value="zh"
+                            value       = "zh",
+                            info        = '识别语言: 自动 (auto)/ 指定'
                         )
-                    with gr.Row():
-                        asr_info = gr.Textbox(label=i18n("ASR进程输出信息"))
-
+                    asr_info = gr.Textbox(label=i18n("ASR 进程输出信息"))
+            
                 def change_lang_choices(key): #根据选择的模型修改可选的语言
                     # return gr.Dropdown(choices=asr_dict[key]['lang'])
-                    return {"__type__": "update", "choices": asr_dict[key]['lang'],"value":asr_dict[key]['lang'][0]}
+                    return {"__type__": "update", "choices": asr_dict[key]['lang'], "value":asr_dict[key]['lang'][0]}
                 def change_size_choices(key): # 根据选择的模型修改可选的模型尺寸
                     # return gr.Dropdown(choices=asr_dict[key]['size'])
-                    return {"__type__": "update", "choices": asr_dict[key]['size']}
+                    if asr_dict[key]['name'] == 'fasterwhisper':
+                        from tools.asr.fasterwhisper_asr import FasterWhisperASR
+                        choice = FasterWhisperASR.check_local_models()
+                    else:
+                        choice = asr_dict[key]['size']
+                    return {"__type__": "update", "choices": choice, "value":choice[0]}
                 asr_model.change(change_lang_choices, [asr_model], [asr_lang])
                 asr_model.change(change_size_choices, [asr_model], [asr_size])
-                
+            
             gr.Markdown(value=i18n("0d-语音文本校对标注工具"))
             with gr.Row():
                 if_label = gr.Checkbox(label=i18n("是否开启打标WebUI"),show_label=True)