-
Notifications
You must be signed in to change notification settings - Fork 707
Replies: 1 comment · 1 reply
-
in general, it requires cuda 11.8 Can you show complete logs? |
Beta Was this translation helpful? Give feedback.
All reactions
-
Yes: Logs
Using 'DEBUG' LOGGING_LEVEL #DEBUG diarize.py [LINE:13] [2025-03-25 18:06:12,414] Diarization_progress: 0.415% Process finished with exit code 0 |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Hello!
I’m using your OfflineSpeakerDiarization tool with CUDA and encountered an issue: GPU usage drops significantly during audio processing. Could you please help me investigate this?
Environment:
CUDA 12.6, sherpa-onnx 1.11.2+cuda
Models:
pyannote-segmentation-3.0 (source: [link])
wespeaker_en_voxceleb_resnet34_LM.onnx (source: [link])
OfflineSpeakerDiarizationConfig parameters: provider="cuda", num_threads=2
Code:
Problem:
When calling model.process(audio_array), GPU usage drops from ~90% to 0%, though GPU memory remains occupied. No errors are logged, but processing becomes noticeably slow:
A 180.0-second audio file with 3 speakers takes 48 seconds with ...resnet34_LM.onnx.
The same file takes 92.9 seconds with wespeaker_en_voxceleb_resnet152_LM.onnx.
Questions:
Is specifying provider="cuda" sufficient and correct for GPU utilization?
Are there any known limitations for CUDA support in speaker diarization?
Which parameters can be adjusted to optimize GPU usage?
Thank you in advance for your help!
Beta Was this translation helpful? Give feedback.
All reactions