search

pyannote / pyannote-audio

Neural building blocks for speaker diarization: speech activity detection, speaker change detection, overlapped speech detection, speaker embedding
http://pyannote.github.io
MIT License
6.38k stars 784 forks source link

Pre-trained models make the tutorial instruction fail #154

Closed hadware closed 5 years ago

hadware commented 5 years ago

This is pretty simple: the pretrained models .pt files make the command

pyannote-speech-detection train --from=0 --to=1000 ${EXPERIMENT_DIR} AMI.SpeakerDiarization.MixHeadset

fail. I've thoroughly tried to tweak the command's options (adding --from option), but i coulnd't get it to run to 1000 epochs without some kind of error.

hbredin commented 5 years ago

Are you referring to the following error message?

FileExistsError: [Errno 17] File exists: 'tutorials/models/speech_activity_detection/train/AMI.SpeakerDiarization.MixHeadset.train/weights'

If so, I will catch the error and make it clear that the directory should be removed first...

hbredin commented 5 years ago

Since e0b73c1, you now get a more explicit error message:

$ pyannote-speech-detection train [REDACTED]/tutorials/models/speech_activity_detection AMI.SpeakerDiarization.MixHeadset
You are about to overwrite pretrained models in "[REDACTED]/tutorials/models/speech_activity_detection/train/AMI.SpeakerDiarization.MixHeadset.train/weights" directory. If you want to train a new model from scratch, first (backup and) remove the directory.

Does this solve this issue?

hadware commented 5 years ago

Yes, that does it. Maybe we could add an option to overwrite all previous weights when running the train script?

hbredin commented 5 years ago

Yes, we could. Not sure it is that helpful though... I'd rather have users remove the directory manually to hold them responsible.

Closing as I think the initial issue is solved.