Speechbrain downloads pretrained models even if local files are passed

[gabrielziegler3]

I noticed that when I ran the example for using ECAPATDNN's pretrained model after downloading the files, speechbrain will read the first two files from filesystem and then will start to request the other files from https://huggingface.co/speechbrain/spkrec-ecapa-voxceleb

To reproduce, run:

import torchaudio
from speechbrain.pretrained import EncoderClassifier
classifier = EncoderClassifier.from_hparams(source="path/to/downloaded/model/dir")
signal, fs =torchaudio.load('samples/audio_samples/example1.wav')
embeddings = classifier.encode_batch(signal)

This behaviour may be seamless for most users, but if you do not have internet in an environment and just wants to load the files from filesystem you will not be able to do so with the from_params method.

I figured what was making speechbrain fallback for requests and I am opening a PR to fix this issue.

[Gastron]

Actually I think the issue is in the hyperparams.yaml file in that particular repo (and unfortunately this pattern may occur elsewhere): https://huggingface.co/speechbrain/spkrec-ecapa-voxceleb/blob/main/hyperparams.yaml#L54-L58

There hyperparams.yaml explicitly specifies the location of the files. This overrides any default_source given to the Pretrainer. There is no need to explicitly specify the location, it would be more flexible if the paths weren't specified. @mravanelli can you remove those lines?

[wengbenjue]

When I load local file,It occur OSError: [Errno 62] Too many levels of symbolic links: 'pretrained_models/EncoderClassifier-3077685261458028297/embedding_model.ckpt',

`
verification = SpeakerRecognition.from_hparams(source="/speaker_verification/tdnn/model/CKPT+2022-04-17+20-14-58+00",savedir="pretrained_models/CKPT+2022-04-17+20-14-58+00")

score, prediction = verification.verify_files("speechbrain/spkrec-ecapa-voxceleb/example1.wav", "speechbrain/spkrec-ecapa-voxceleb/example2.flac")
print(score,prediction)
score, prediction = verification.verify_files("speechbrain/spkrec-ecapa-voxceleb/example1.wav", "speechbrain/spkrec-ecapa-voxceleb/example1.wav")
print(score,prediction)
`

[Adel-Moumen]

Hello,

Any news regarding this issue please? Thanks.

Adel