bbrookie
commented
3 years ago Sorry,Until now, this kind of problem will still appear after I add the model again. I don't have any solution. Can you help me answer it? the traceback like this:
Traceback (most recent call last): File "/home/zhongcy/Public/pyannote-audio-master/notebooks/demo.py", line 28, in <module> method=method) File "/home/zhongcy/anaconda3/envs/audio/lib/python3.7/site-packages/pyannote/audio/pipeline/speaker_diarization.py", line 132, in __init__ embedding=self.embedding, metric=self.metric, method=self.method File "/home/zhongcy/anaconda3/envs/audio/lib/python3.7/site-packages/pyannote/audio/pipeline/speech_turn_clustering.py", line 78, in __init__ self._embedding = Wrapper(self.embedding) File "/home/zhongcy/anaconda3/envs/audio/lib/python3.7/site-packages/pyannote/audio/features/wrapper.py", line 217, in __init__ raise ValueError(msg) ValueError: "/home/zhongcy/Public/my_model/models/emb_resnet/train/X.SpeakerDiarization.VoxCeleb.train/weights/0326.pt" directory does not seem to be the path to a pretrained model checkpoint.
I used another independent models to train embedding. Now I added the network to pyannote. The final file structure is set to be the same as the file structure obtained by pyannote training（The specific operation is that I directly copied all the files of emb_voxceleb in pyannote-audio-hub, just replaced my model file and changed the name of my model to {epoch}.pt, and modified the network parameters in the config.yml file ）
but when I use --pretrained to specify the checkpoint path , Will raise an error:
directory does not seem to be the path to a pretrained model checkpoint.I tried to do this before, and the result is feasible (Although it will report the same error ，I didn’t make any special changes, it can be used after a while), but if a new model is added, the same error will still occur， this time I want to know how to solve this problem, so that I can directly use the network I trained separately
Looking forward to your reply ， thank you