Give an informative log for whisper on exceptions. (#473)

python-api-examples/generate-subtitles.py

@@ -180,6 +180,17 @@ def get_args():
         """,
     )
 
+    parser.add_argument(
+        "--whisper-tail-paddings",
+        default=-1,
+        type=int,
+        help="""Number of tail padding frames.
+        We have removed the 30-second constraint from whisper, so you need to
+        choose the amount of tail padding frames by yourself.
+        Use -1 to use a default value for tail padding.
+        """,
+    )
+
     parser.add_argument(
         "--decoding-method",
         type=str,
@@ -294,6 +305,7 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
             debug=args.debug,
             language=args.whisper_language,
             task=args.whisper_task,
+            tail_paddings=args.whisper_tail_paddings,
         )
     else:
         raise ValueError("Please specify at least one model")

python-api-examples/non_streaming_server.py

@@ -277,6 +277,17 @@ def add_whisper_model_args(parser: argparse.ArgumentParser):
         """,
     )
 
+    parser.add_argument(
+        "--whisper-tail-paddings",
+        default=-1,
+        type=int,
+        help="""Number of tail padding frames.
+        We have removed the 30-second constraint from whisper, so you need to
+        choose the amount of tail padding frames by yourself.
+        Use -1 to use a default value for tail padding.
+        """,
+    )
+
 
 def add_model_args(parser: argparse.ArgumentParser):
     add_transducer_model_args(parser)
@@ -913,6 +924,7 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
             decoding_method=args.decoding_method,
             language=args.whisper_language,
             task=args.whisper_task,
+            tail_paddings=args.whisper_tail_paddings,
         )
     elif args.tdnn_model:
         assert_file_exists(args.tdnn_model)

python-api-examples/offline-decode-files.py

@@ -220,6 +220,17 @@ def get_args():
         """,
     )
 
+    parser.add_argument(
+        "--whisper-tail-paddings",
+        default=-1,
+        type=int,
+        help="""Number of tail padding frames.
+        We have removed the 30-second constraint from whisper, so you need to
+        choose the amount of tail padding frames by yourself.
+        Use -1 to use a default value for tail padding.
+        """,
+    )
+
     parser.add_argument(
         "--decoding-method",
         type=str,
@@ -391,6 +402,7 @@ def main():
             debug=args.debug,
             language=args.whisper_language,
             task=args.whisper_task,
+            tail_paddings=args.whisper_tail_paddings,
         )
     elif args.tdnn_model:
         assert_file_exists(args.tdnn_model)

python-api-examples/two-pass-speech-recognition-from-microphone.py

@@ -195,6 +195,17 @@ def add_second_pass_whisper_model_args(parser: argparse.ArgumentParser):
         """,
     )
 
+    parser.add_argument(
+        "--second-whisper-tail-paddings",
+        default=-1,
+        type=int,
+        help="""Number of tail padding frames.
+        We have removed the 30-second constraint from whisper, so you need to
+        choose the amount of tail padding frames by yourself.
+        Use -1 to use a default value for tail padding.
+        """,
+    )
+
 
 def add_second_pass_non_streaming_model_args(parser: argparse.ArgumentParser):
     add_second_pass_transducer_model_args(parser)
@@ -314,6 +325,7 @@ def create_second_pass_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
             decoding_method="greedy_search",
             language=args.second_whisper_language,
             task=args.second_whisper_task,
+            tail_paddings=args.second_whisper_tail_paddings,
         )
     else:
         raise ValueError("Please specify at least one model for the second pass")

python-api-examples/vad-with-non-streaming-asr.py

@@ -166,6 +166,17 @@ def get_args():
         """,
     )
 
+    parser.add_argument(
+        "--whisper-tail-paddings",
+        default=-1,
+        type=int,
+        help="""Number of tail padding frames.
+        We have removed the 30-second constraint from whisper, so you need to
+        choose the amount of tail padding frames by yourself.
+        Use -1 to use a default value for tail padding.
+        """,
+    )
+
     parser.add_argument(
         "--decoding-method",
         type=str,
@@ -256,6 +267,7 @@ def create_recognizer(args) -> sherpa_onnx.OfflineRecognizer:
             debug=args.debug,
             language=args.whisper_language,
             task=args.whisper_task,
+            tail_paddings=args.whisper_tail_paddings,
         )
     else:
         raise ValueError("Please specify at least one model")