Add fetch local_strategy parameter and disable symlinks by default

recipes/ESC50/esc50_prepare.py

@@ -21,7 +21,7 @@
 
 import speechbrain as sb
 from speechbrain.dataio.dataio import load_data_csv, read_audio
-from speechbrain.utils.fetching import fetch
+from speechbrain.utils.fetching import LocalStrategy, fetch
 
 logger = logging.getLogger(__name__)
 
@@ -47,14 +47,16 @@ def download_esc50(data_path):
         temp_path = os.path.join(data_path, "temp_download")
 
         # download the data
-        fetch(
+        archive_path = fetch(
             "master.zip",
             "https://github.com/karoldvl/ESC-50/archive/",
             savedir=temp_path,
+            # URL, so will be fetched directly in the savedir anyway
+            local_strategy=LocalStrategy.COPY_SKIP_CACHE,
         )
 
         # unpack the .zip file
-        shutil.unpack_archive(os.path.join(temp_path, "master.zip"), data_path)
+        shutil.unpack_archive(archive_path, data_path)
 
         # move the files up to the datapath
         files = os.listdir(os.path.join(data_path, "ESC-50-master"))

speechbrain/inference/classifiers.py

@@ -20,7 +20,7 @@
 import speechbrain
 from speechbrain.inference.interfaces import Pretrained
 from speechbrain.utils.data_utils import split_path
-from speechbrain.utils.fetching import fetch
+from speechbrain.utils.fetching import LocalStrategy, fetch
 
 
 class EncoderClassifier(Pretrained):
@@ -293,7 +293,12 @@ def classify_file(self, path, savedir="audio_cache"):
             (label encoder should be provided).
         """
         source, fl = split_path(path)
-        path = fetch(fl, source=source, savedir=savedir)
+        path = fetch(
+            fl,
+            source=source,
+            savedir=savedir,
+            local_strategy=LocalStrategy.NO_LINK,
+        )
 
         batch, fs_file = torchaudio.load(path)
         batch = batch.to(self.device)

speechbrain/inference/interfaces.py

@@ -32,7 +32,7 @@
 from speechbrain.utils.data_pipeline import DataPipeline
 from speechbrain.utils.data_utils import split_path
 from speechbrain.utils.distributed import run_on_main
-from speechbrain.utils.fetching import fetch
+from speechbrain.utils.fetching import LocalStrategy, fetch
 from speechbrain.utils.superpowers import import_from_path
 
 logger = logging.getLogger(__name__)
@@ -49,6 +49,7 @@ def foreign_class(
     use_auth_token=False,
     download_only=False,
     huggingface_cache_dir=None,
+    local_strategy: LocalStrategy = LocalStrategy.NO_LINK,
     **kwargs,
 ):
     """Fetch and load an interface from an outside source
@@ -95,6 +96,10 @@ def foreign_class(
         If true, class and instance creation is skipped.
     huggingface_cache_dir : str
         Path to HuggingFace cache; if None -> "~/.cache/huggingface" (default: None)
+    local_strategy : speechbrain.utils.fetching.LocalStrategy
+        The fetching strategy to use, which controls the behavior of remote file
+        fetching with regards to symlinking and copying.
+        See :func:`speechbrain.utils.fetching.fetch` for further details.
     **kwargs : dict
         Arguments to forward to class constructor.
 
@@ -114,6 +119,7 @@ def foreign_class(
         use_auth_token=use_auth_token,
         revision=None,
         huggingface_cache_dir=huggingface_cache_dir,
+        local_strategy=local_strategy,
     )
     pymodule_local_path = fetch(
         filename=pymodule_file,
@@ -124,6 +130,7 @@ def foreign_class(
         use_auth_token=use_auth_token,
         revision=None,
         huggingface_cache_dir=huggingface_cache_dir,
+        local_strategy=local_strategy,
     )
     sys.path.append(str(pymodule_local_path.parent))
 
@@ -286,7 +293,12 @@ def load_audio(self, path, savedir="."):
         The path can be a local path, a web url, or a link to a huggingface repo.
         """
         source, fl = split_path(path)
-        path = fetch(fl, source=source, savedir=savedir)
+        path = fetch(
+            fl,
+            source=source,
+            savedir=savedir,
+            local_strategy=LocalStrategy.NO_LINK,
+        )
         signal, sr = torchaudio.load(str(path), channels_first=False)
         return self.audio_normalizer(signal, sr)
 
@@ -397,6 +409,7 @@ def from_hparams(
         download_only=False,
         huggingface_cache_dir=None,
         overrides_must_match=True,
+        local_strategy: LocalStrategy = LocalStrategy.NO_LINK,
         **kwargs,
     ):
         """Fetch and load based from outside source based on HyperPyYAML file
@@ -450,6 +463,9 @@ def from_hparams(
             Path to HuggingFace cache; if None -> "~/.cache/huggingface" (default: None)
         overrides_must_match : bool
             Whether the overrides must match the parameters already in the file.
+        local_strategy : LocalStrategy, optional
+            Which strategy to use to deal with files locally. (default:
+            `LocalStrategy.SYMLINK`)
         **kwargs : dict
             Arguments to forward to class constructor.
 
@@ -469,6 +485,7 @@ def from_hparams(
             use_auth_token=use_auth_token,
             revision=revision,
             huggingface_cache_dir=huggingface_cache_dir,
+            local_strategy=local_strategy,
         )
         try:
             pymodule_local_path = fetch(
@@ -480,6 +497,7 @@ def from_hparams(
                 use_auth_token=use_auth_token,
                 revision=revision,
                 huggingface_cache_dir=huggingface_cache_dir,
+                local_strategy=local_strategy,
             )
             sys.path.append(str(pymodule_local_path.parent))
         except ValueError:

speechbrain/inference/interpretability.py

@@ -22,7 +22,7 @@
 from speechbrain.inference.interfaces import Pretrained
 from speechbrain.processing.NMF import spectral_phase
 from speechbrain.utils.data_utils import split_path
-from speechbrain.utils.fetching import fetch
+from speechbrain.utils.fetching import LocalStrategy, fetch
 
 
 class PIQAudioInterpreter(Pretrained):
@@ -153,7 +153,12 @@ def interpret_file(self, path, savedir="audio_cache"):
             The sampling frequency of the model. Useful to save the audio.
         """
         source, fl = split_path(path)
-        path = fetch(fl, source=source, savedir=savedir)
+        path = fetch(
+            fl,
+            source=source,
+            savedir=savedir,
+            local_strategy=LocalStrategy.NO_LINK,
+        )
 
         batch, fs_file = torchaudio.load(path)
         batch = batch.to(self.device)

speechbrain/inference/separation.py

@@ -20,7 +20,7 @@
 
 from speechbrain.inference.interfaces import Pretrained
 from speechbrain.utils.data_utils import split_path
-from speechbrain.utils.fetching import fetch
+from speechbrain.utils.fetching import LocalStrategy, fetch
 
 
 class SepformerSeparation(Pretrained):
@@ -97,7 +97,12 @@ def separate_file(self, path, savedir="audio_cache"):
             Separated sources
         """
         source, fl = split_path(path)
-        path = fetch(fl, source=source, savedir=savedir)
+        path = fetch(
+            fl,
+            source=source,
+            savedir=savedir,
+            local_strategy=LocalStrategy.NO_LINK,
+        )
 
         batch, fs_file = torchaudio.load(path)
         batch = batch.to(self.device)