Fix timestamp extraction bug in transcribe function

This commit is contained in:
crackerben 2023-12-06 20:34:30 +08:00
parent 0bf5d0e3bc
commit 6cbc029450

View File

@ -300,7 +300,7 @@ def transcribe(
seek += last_timestamp_pos * input_stride seek += last_timestamp_pos * input_stride
else: else:
duration = segment_duration duration = segment_duration
timestamps = tokens[timestamp_tokens.nonzero().flatten()] timestamps = tokens[timestamp_tokens.nonzero()[0]]
if ( if (
len(timestamps) > 0 len(timestamps) > 0
and timestamps[-1].item() != tokenizer.timestamp_begin and timestamps[-1].item() != tokenizer.timestamp_begin