生成 rnnt eou 训练数据集

readme.txt

@@ -243,6 +243,13 @@ python examples/asr/asr_cache_aware_streaming/speech_to_text_cache_aware_streami
 
 
 
+生成 rnnt eou 数据集
+python sou_eou_time.py
+    复用前面 CTC 生成的数据，只是加字段
+
+
+
+
 它是不是真的流式输入，流式输出，为什么它是一下子打印全部结果，而不是边推理边输出部分结果 `examples\asr\asr_cache_aware_streaming\speech_to_text_cache_aware_streaming_infer.py` 
    
 它**确实是真流式**的（前提是你用了支持 Cache-aware 的模型并搭配这个脚本）。

sou_eou_time.py

@@ -0,0 +1,56 @@
+import json
+import os
+
+def process_manifest(input_path):
+    if not os.path.exists(input_path):
+        print(f"Warning: File not found {input_path}")
+        return
+
+    print(f"Processing {input_path} in-place")
+    
+    processed_count = 0
+    updated_lines = []
+    
+    # 1. Read and process all lines
+    with open(input_path, 'r', encoding='utf-8') as f_in:
+        for line in f_in:
+            line = line.strip()
+            if not line:
+                continue
+                
+            data = json.loads(line)
+            
+            if 'duration' not in data:
+                print(f"Warning: 'duration' missing in a record, skipping: {data}")
+                updated_lines.append(line + '\n')
+                continue
+                
+            duration = float(data['duration'])
+            
+            if 'offset' not in data:
+                data['offset'] = 0.0
+                
+            if 'sou_time' not in data:
+                data['sou_time'] = 0.0
+                
+            if 'eou_time' not in data:
+                data['eou_time'] = duration
+                
+            updated_lines.append(json.dumps(data, ensure_ascii=False) + '\n')
+            processed_count += 1
+            
+    # 2. Write back to the same file
+    with open(input_path, 'w', encoding='utf-8') as f_out:
+        f_out.writelines(updated_lines)
+            
+    print(f"Successfully processed {processed_count} records in {input_path}.")
+
+if __name__ == '__main__':
+    files_to_process = [
+        r"data\common_voice_11_0\ja\train_tarred_1bk\tarred_audio_manifest.json",
+        r"data\common_voice_11_0\ja\validation\validation_common_voice_11_0_manifest.json",
+        r"data\common_voice_11_0\ja\test\test_common_voice_11_0_manifest.json"
+    ]
+    
+    for file_path in files_to_process:
+        process_manifest(file_path)