huggingface_hub.hf_hub_download supports passing a revision parameter to hf_hub_{url|download} to load a specific version (understand: commit, tag, or branch head) of models hosted on Huggingface Hub:

from huggingface_hub import hf_hub_download
hf_hub_url(repo_id, filename, revision=..., ...)

It would be nice to expose this revision parameter in speechbrain so that one can pin a model to a specific version (and ensure that a given code will still behave the same even if the model is updated upstream).

The API could look like this:

from speechbrain.pretrained import EncoderClassifier
encoder = EncoderClassifier.from_hparams(
    source="speechbrain/spkrec-ecapa-voxceleb",
    revision="5c0be3875fda05e81f3c004ed8c7c06be308de1e",
)

or maybe like this that @julien-c implemented for pyannote.audio:

from speechbrain.pretrained import EncoderClassifier
encoder = EncoderClassifier.from_hparams(
    source="speechbrain/spkrec-ecapa-voxceleb@5c0be3875fda05e81f3c004ed8c7c06be308de1e")