Update app.py

app.py

@@ -25,10 +25,6 @@ def combine_audio_with_time(target_audio, mixed_audio):
     if pipeline is None:
         return "错误: 模型未初始化"
     
-    # 打印文件路径，确保文件正确传递
-    print(f"目标音频文件路径: {target_audio}")
-    print(f"混合音频文件路径: {mixed_audio}")
-
     # 加载目标说话人的样本音频
     try:
         target_audio_segment = AudioSegment.from_wav(target_audio)
@@ -68,31 +64,43 @@ def diarize_audio(temp_file):
     # 返回 diarization 输出
     return str(diarization)
 
-# 生成标签文件的函数
-def generate_labels_from_diarization(diarization_output):
-    labels_path = 'labels.txt'
-    successful_lines = 0
-
-    try:
-        with open(labels_path, 'w') as outfile:
-            lines = diarization_output.strip().split('\n')
-            for line in lines:
-                try:
-                    parts = line.strip()[1:-1].split(' --> ')
-                    start_time = parts[0].strip()
-                    end_time = parts[1].split(']')[0].strip()
-                    label = line.split()[-1].strip()
-                    start_seconds = timestamp_to_seconds(start_time)
-                    end_seconds = timestamp_to_seconds(end_time)
-                    outfile.write(f"{start_seconds}\t{end_seconds}\t{label}\n")
-                    successful_lines += 1
-                except Exception as e:
-                    print(f"处理行时出错: '{line.strip()}'. 错误: {e}")
-        print(f"成功处理了 {successful_lines} 行。")
-        return labels_path if successful_lines > 0 else None
-    except Exception as e:
-        print(f"写入文件时出错: {e}")
-        return None
+# 计算时间段的重叠部分（单位：秒）
+def calculate_overlap(start1, end1, start2, end2):
+    overlap_start = max(start1, start2)
+    overlap_end = min(end1, end2)
+    overlap_duration = max(0, overlap_end - overlap_start)
+    return overlap_duration
+
+# 获取目标时间段和说话人时间段的重叠比例
+def get_best_match(target_time, diarization_output):
+    target_start_time = target_time['start_time']
+    target_end_time = target_time['end_time']
+    
+    # 假设 diarization_output 是一个列表，包含说话人时间段和标签
+    speaker_segments = []
+    for line in diarization_output.strip().split('\n'):
+        try:
+            parts = line.strip()[1:-1].split(' --> ')
+            start_time = parts[0].strip()
+            end_time = parts[1].split(']')[0].strip()
+            label = line.split()[-1].strip()
+
+            start_seconds = timestamp_to_seconds(start_time)
+            end_seconds = timestamp_to_seconds(end_time)
+
+            # 计算目标音频时间段和说话人时间段的重叠时间
+            overlap = calculate_overlap(target_start_time, target_end_time, start_seconds, end_seconds)
+            overlap_ratio = overlap / (end_seconds - start_seconds)
+
+            # 记录说话人标签和重叠比例
+            speaker_segments.append((label, overlap_ratio, start_seconds, end_seconds))
+
+        except Exception as e:
+            print(f"处理行时出错: '{line.strip()}'. 错误: {e}")
+    
+    # 按照重叠比例排序，返回重叠比例最大的一段
+    best_match = max(speaker_segments, key=lambda x: x[1], default=None)
+    return best_match
 
 # 将时间戳转换为秒
 def timestamp_to_seconds(timestamp):
@@ -105,9 +113,6 @@ def timestamp_to_seconds(timestamp):
 
 # 处理音频文件并返回输出
 def process_audio(target_audio, mixed_audio):
-    # 打印文件路径，确保传入的文件有效
-    print(f"处理音频：目标音频: {target_audio}, 混合音频: {mixed_audio}")
-    
     # 进行音频拼接并返回目标音频的起始和结束时间（作为字典）
     time_dict = combine_audio_with_time(target_audio, mixed_audio)
     
@@ -117,32 +122,30 @@ def process_audio(target_audio, mixed_audio):
     if diarization_result.startswith("错误"):
         return diarization_result, None, None  # 出错时返回错误信息
     else:
-        # 生成标签文件
-        label_file = generate_labels_from_diarization(diarization_result)
-        return diarization_result, label_file, time_dict  # 返回说话人分离结果、标签文件和目标音频的时间段
+        # 获取最佳匹配的说话人时间段
+        best_match = get_best_match(time_dict, diarization_result)
+        return diarization_result, best_match  # 返回说话人分离结果和最佳匹配的说话人时间段
 
 # Gradio 接口
 with gr.Blocks() as demo:
     gr.Markdown("""
     # 🗣️ 音频拼接与说话人分类 🗣️
-    上传目标说话人音频和混合音频，拼接并进行说话人分类。结果包括说话人分离输出、标签文件和目标音频的时间段。
+    上传目标音频和混合音频，拼接并进行说话人分类。结果包括说话人分离输出、最佳匹配说话人时间段。
     """)
     mixed_audio_input = gr.Audio(type="filepath", label="上传混合音频")
     target_audio_input = gr.Audio(type="filepath", label="上传目标说话人音频")
     
-    
     process_button = gr.Button("处理音频")
     
     # 输出结果
     diarization_output = gr.Textbox(label="说话人分离结果")
-    label_file_link = gr.File(label="下载标签文件")
-    time_range_output = gr.Textbox(label="目标音频时间段")
+    best_match_output = gr.Textbox(label="最佳匹配说话人时间段")
 
     # 点击按钮时触发处理音频
     process_button.click(
         fn=process_audio,
         inputs=[target_audio_input, mixed_audio_input],
-        outputs=[diarization_output, label_file_link, time_range_output]
+        outputs=[diarization_output, best_match_output]
     )
 
 demo.launch(share=True)