krugd/dnd_transcribe
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Allow partial transcription when streaming

Browse Source
This commit is contained in:
David Kruger 2025-04-28 00:09:00 -07:00
parent db2b6d956b
commit 72347d5d47
1 changed files with 12 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
dnd_transcribe/inference.py
View File

@ -62,14 +62,18 @@ class Inference:
for block in stream: for block in stream:
if len(block.shape) > 1: if len(block.shape) > 1:
block = speech[:, 0] + speech[:, 1] block = speech[:, 0] + speech[:, 1]
block_inference = self.buffer_to_text(block) try:
transcript += block_inference.transcript + " " block_inference = self.buffer_to_text(block)
processing_time += block_inference.processing_time_sec transcript += block_inference.transcript + " "
if block_inference.confidence_score is not None: processing_time += block_inference.processing_time_sec
if confidence is None: if block_inference.confidence_score is not None:
confidence = block_inference.confidence_score if confidence is None:
else: confidence = block_inference.confidence_score
confidence *= block_inference.confidence_score else:
confidence *= block_inference.confidence_score
except torch.OutOfMemoryError as e:
print(e)
break
return InferredTranscript(transcript.strip(), confidence, processing_time) return InferredTranscript(transcript.strip(), confidence, processing_time)
def file_to_text(self, audio_file: typing.BinaryIO) -> InferredTranscript: def file_to_text(self, audio_file: typing.BinaryIO) -> InferredTranscript: