nkululeko 0.82.2__tar.gz → 0.82.4__tar.gz

{nkululeko-0.82.2 → nkululeko-0.82.4}/CHANGELOG.md

@@ -1,6 +1,14 @@
 Changelog
 =========
 
+Version 0.82.4
+--------------
+* fixed bug in wavlm 
+
+Version 0.82.3
+--------------
+* fixed another audformat peculiarity to interprete time values as nanoseconds
+
 Version 0.82.2
 --------------
 * fixed audformat peculiarity that dataframes can have only one column

{nkululeko-0.82.2/nkululeko.egg-info → nkululeko-0.82.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nkululeko
-Version: 0.82.2
+Version: 0.82.4
 Summary: Machine learning audio prediction experiments based on templates
 Home-page: https://github.com/felixbur/nkululeko
 Author: Felix Burkhardt
@@ -333,6 +333,14 @@ F. Burkhardt, Johannes Wagner, Hagen Wierstorf, Florian Eyben and Björn Schulle
 Changelog
 =========
 
+Version 0.82.4
+--------------
+* fixed bug in wavlm 
+
+Version 0.82.3
+--------------
+* fixed another audformat peculiarity to interprete time values as nanoseconds
+
 Version 0.82.2
 --------------
 * fixed audformat peculiarity that dataframes can have only one column

{nkululeko-0.82.2 → nkululeko-0.82.4}/nkululeko/constants.py

@@ -1,2 +1,2 @@
-VERSION="0.82.2"
+VERSION="0.82.4"
 SAMPLING_RATE = 16000

{nkululeko-0.82.2 → nkululeko-0.82.4}/nkululeko/data/dataset_csv.py

@@ -22,7 +22,16 @@ class Dataset_CSV(Dataset):
         #     data_file = os.path.join(exp_root, data_file)
         root = os.path.dirname(data_file)
         audio_path = self.util.config_val_data(self.name, "audio_path", "")
-        df = audformat.utils.read_csv(data_file)
+        df = pd.read_csv(data_file)
+        # special treatment for segmented dataframes with only one column:
+        if "start" in df.columns and len(df.columns) == 4:
+            index = audformat.segmented_index(
+                df.file.values, df.start.values, df.end.values
+            )
+            df = df.set_index(index)
+            df = df.drop(columns=["file", "start", "end"])
+        else:
+            df = audformat.utils.read_csv(data_file)
         if isinstance(df, pd.Series):
             df = df.to_frame()
         rename_cols = self.util.config_val_data(self.name, "colnames", False)

{nkululeko-0.82.2 → nkululeko-0.82.4}/nkululeko/feat_extract/feats_hubert.py

@@ -6,23 +6,26 @@
 
 import os
 
-import audeer
-import nkululeko.glob_conf as glob_conf
 import pandas as pd
 import torch
 import torchaudio
-from audformat.utils import map_file_path
-from nkululeko.feat_extract.featureset import Featureset
 from tqdm import tqdm
-from transformers import HubertModel, Wav2Vec2FeatureExtractor
+from transformers import HubertModel
+from transformers import Wav2Vec2FeatureExtractor
+
+from nkululeko.feat_extract.featureset import Featureset
+import nkululeko.glob_conf as glob_conf
 
 
 class Hubert(Featureset):
-    """Class to extract HuBERT embedding)"""
+    """Class to extract HuBERT embedding)."""
 
     def __init__(self, name, data_df, feat_type):
-        """Constructor. is_train is needed to distinguish from test/dev sets,
-        because they use the codebook from the training"""
+        """Constructor.
+
+        Is_train is needed to distinguish from test/dev sets,
+        because they use the codebook from the training.
+        """
         super().__init__(name, data_df, feat_type)
         # check if device is not set, use cuda if available
         cuda = "cuda" if torch.cuda.is_available() else "cpu"
@@ -61,16 +64,12 @@ class Hubert(Featureset):
         """Extract the features or load them from disk if present."""
         store = self.util.get_path("store")
         storage = f"{store}{self.name}.pkl"
-        extract = self.util.config_val(
-            "FEATS", "needs_feature_extraction", False
-        )
+        extract = self.util.config_val("FEATS", "needs_feature_extraction", False)
         no_reuse = eval(self.util.config_val("FEATS", "no_reuse", "False"))
         if extract or no_reuse or not os.path.isfile(storage):
             if not self.model_initialized:
                 self.init_model()
-            self.util.debug(
-                "extracting Hubert embeddings, this might take a while..."
-            )
+            self.util.debug("extracting Hubert embeddings, this might take a while...")
             emb_series = pd.Series(index=self.data_df.index, dtype=object)
             length = len(self.data_df.index)
             for idx, (file, start, end) in enumerate(
@@ -84,9 +83,7 @@ class Hubert(Featureset):
                 assert sampling_rate == 16000
                 emb = self.get_embeddings(signal, sampling_rate, file)
                 emb_series.iloc[idx] = emb
-            self.df = pd.DataFrame(
-                emb_series.values.tolist(), index=self.data_df.index
-            )
+            self.df = pd.DataFrame(emb_series.values.tolist(), index=self.data_df.index)
             self.df.to_pickle(storage)
             try:
                 glob_conf.config["DATA"]["needs_feature_extraction"] = "false"

{nkululeko-0.82.2 → nkululeko-0.82.4}/nkululeko/feat_extract/feats_wavlm.py

@@ -4,27 +4,32 @@
 
 import os
 
-import nkululeko.glob_conf as glob_conf
 import pandas as pd
 import torch
 import torchaudio
-from nkululeko.feat_extract.featureset import Featureset
 from tqdm import tqdm
-from transformers import Wav2Vec2FeatureExtractor, WavLMModel
+from transformers import Wav2Vec2FeatureExtractor
+from transformers import WavLMModel
+
+from nkululeko.feat_extract.featureset import Featureset
+import nkululeko.glob_conf as glob_conf
 
 
 class Wavlm(Featureset):
-    """Class to extract WavLM embedding)"""
+    """Class to extract WavLM embedding)."""
+
+    def __init__(self, name, data_df, feats_type):
+        """Constructor.
 
-    def __init__(self, name, data_df, feat_type):
-        """Constructor. is_train is needed to distinguish from test/dev sets,
-        because they use the codebook from the training"""
-        super().__init__(name, data_df)
+        Is_train is needed to distinguish from test/dev sets,
+        because they use the codebook from the training.
+        """
+        super().__init__(name, data_df, feats_type)
         # check if device is not set, use cuda if available
         cuda = "cuda" if torch.cuda.is_available() else "cpu"
         self.device = self.util.config_val("MODEL", "device", cuda)
         self.model_initialized = False
-        self.feat_type = feat_type
+        self.feat_type = feats_type
 
     def init_model(self):
         # load model
@@ -59,7 +64,9 @@ class Wavlm(Featureset):
                     frame_offset=int(start.total_seconds() * 16000),
                     num_frames=int((end - start).total_seconds() * 16000),
                 )
-                assert sampling_rate == 16000, f"sampling rate should be 16000 but is {sampling_rate}"
+                assert (
+                    sampling_rate == 16000
+                ), f"sampling rate should be 16000 but is {sampling_rate}"
                 emb = self.get_embeddings(signal, sampling_rate, file)
                 emb_series.iloc[idx] = emb
             self.df = pd.DataFrame(emb_series.values.tolist(), index=self.data_df.index)

{nkululeko-0.82.2 → nkululeko-0.82.4/nkululeko.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nkululeko
-Version: 0.82.2
+Version: 0.82.4
 Summary: Machine learning audio prediction experiments based on templates
 Home-page: https://github.com/felixbur/nkululeko
 Author: Felix Burkhardt
@@ -333,6 +333,14 @@ F. Burkhardt, Johannes Wagner, Hagen Wierstorf, Florian Eyben and Björn Schulle
 Changelog
 =========
 
+Version 0.82.4
+--------------
+* fixed bug in wavlm 
+
+Version 0.82.3
+--------------
+* fixed another audformat peculiarity to interprete time values as nanoseconds
+
 Version 0.82.2
 --------------
 * fixed audformat peculiarity that dataframes can have only one column