Upload app.py

app.py

@@ -2,8 +2,8 @@ import gradio as gr
 import numpy as np
 import torch
 import torchaudio
+import re
 from funasr import AutoModel
-from funasr.utils.postprocess_utils import rich_transcription_postprocess
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 model = AutoModel(
@@ -76,7 +76,14 @@ emo_set = {"😊", "😔", "😡", "😰", "🤢", "😮"}
 event_set = {"🎼", "👏", "😀", "😭", "🤧", "😷"}
 
 
+def clean_text_postprocess(raw_text):
+    """彻底剥离所有模型内部标签，输出绝对纯净的文字（绝不含任何 Emoji）"""
+    return re.sub(r"<\|[^>]*\|?>", "", raw_text).strip()
+
+
 def format_to_emoji(raw_text):
+    """将模型标签转换为 Emoji 表情格式"""
+
     def format_part(s):
         sptk_dict = {}
         for sptk in emoji_dict:
@@ -136,7 +143,7 @@ def generate_srt(sentence_info, clean=True):
         end = ms_to_srt_time(seg.get("end", 0))
         text = seg.get("text", "")
         if clean:
-            text = rich_transcription_postprocess(text)
+            text = clean_text_postprocess(text)
         text = text.strip()
         if text:
             srt_lines.append(f"{i}\n{start} --> {end}\n{text}\n")
@@ -177,14 +184,14 @@ def model_inference(
     sentence_info = res[0].get("sentence_info", [])
 
     if output_format == "纯净文本":
-        return rich_transcription_postprocess(raw_text)
+        return clean_text_postprocess(raw_text)
     elif output_format == "原始富文本":
         return raw_text
     elif output_format == "Emoji 格式":
         return format_to_emoji(raw_text)
     elif output_format == "SRT 字幕":
         srt_result = generate_srt(sentence_info, clean=True)
-        return srt_result if srt_result else rich_transcription_postprocess(raw_text)
+        return srt_result if srt_result else clean_text_postprocess(raw_text)
     elif output_format == "MASTER_DATA":
         srt_raw = generate_srt(sentence_info, clean=False)
         return srt_raw if srt_raw else raw_text
@@ -196,19 +203,24 @@ custom_css = """
 body, .gradio-container {
     font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Oxygen, Ubuntu, Cantarell, 'Open Sans', 'Helvetica Neue', sans-serif;
 }
-div[data-testid="dropdown"],
-div[data-testid="dropdown"] * {
+.custom-dropdown,
+.custom-dropdown *,
+.custom-dropdown input,
+.custom-dropdown .wrap,
+.custom-dropdown .wrap-inner {
     cursor: pointer !important;
 }
-div[data-testid="dropdown"] svg,
-.dropdown-arrow {
-    transition: transform 0.2s ease-in-out !important;
-    transform-origin: 50% 50% !important;
+.custom-dropdown svg,
+.custom-dropdown .dropdown-arrow,
+.custom-dropdown .secondary-wrap svg {
+    transition: transform 0.25s ease-in-out !important;
+    transform-origin: center center !important;
 }
-div[data-testid="dropdown"]:focus-within svg,
-div[data-testid="dropdown"]:has(ul) svg,
-div[data-testid="dropdown"]:has(.options) svg,
-div[data-testid="dropdown"][aria-expanded="true"] svg {
+.custom-dropdown:focus-within svg,
+.custom-dropdown:has(input:focus) svg,
+.custom-dropdown:has(.options) svg,
+.custom-dropdown:has(ul) svg,
+.custom-dropdown input:focus ~ .secondary-wrap svg {
     transform: rotate(180deg) !important;
 }
 """
@@ -228,12 +240,16 @@ with gr.Blocks(theme=gr.themes.Soft(), css=custom_css) as demo:
         with gr.Column(scale=2):
             audio_inputs = gr.Audio(label="上传/录制音频")
             language_inputs = gr.Dropdown(
-                choices=lang_options, value="auto", label="源语言"
+                choices=lang_options,
+                value="auto",
+                label="源语言",
+                elem_classes=["custom-dropdown"],
             )
             output_format_dropdown = gr.Dropdown(
                 choices=["纯净文本", "原始富文本", "Emoji 格式", "SRT 字幕"],
                 value="纯净文本",
                 label="输出格式",
+                elem_classes=["custom-dropdown"],
             )
             with gr.Accordion("高级设置", open=False):
                 use_itn_checkbox = gr.Checkbox(value=True, label="自动添加标点与格式化")