nkululeko 0.60.0__py3-none-any.whl → 0.61.0__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (9) hide show
  1. nkululeko/constants.py +1 -1
  2. nkululeko/feature_extractor.py +90 -47
  3. nkululeko/plots.py +11 -0
  4. nkululeko/segment.py +23 -27
  5. {nkululeko-0.60.0.dist-info → nkululeko-0.61.0.dist-info}/METADATA +5 -1
  6. {nkululeko-0.60.0.dist-info → nkululeko-0.61.0.dist-info}/RECORD +9 -9
  7. {nkululeko-0.60.0.dist-info → nkululeko-0.61.0.dist-info}/LICENSE +0 -0
  8. {nkululeko-0.60.0.dist-info → nkululeko-0.61.0.dist-info}/WHEEL +0 -0
  9. {nkululeko-0.60.0.dist-info → nkululeko-0.61.0.dist-info}/top_level.txt +0 -0
nkululeko/constants.py CHANGED
@@ -1 +1 @@
1
- VERSION = '0.60.0'
1
+ VERSION = '0.61.0'
nkululeko/feature_extractor.py CHANGED
@@ -5,8 +5,9 @@ Helper class to encapsulate feature extraction methods
5
5
 
6
6
  """
7
7
  import pandas as pd
8
- from nkululeko.util import Util
8
+
9
9
  from nkululeko.feat_extract.feats_opensmile import Opensmileset
10
+ from nkululeko.util import Util
10
11
 
11
12
 
12
13
  class FeatureExtractor:
@@ -16,80 +17,122 @@ class FeatureExtractor:
16
17
  data_df (pandas.DataFrame): dataframe with audiofile paths as index
17
18
  feats_types (array of strings): designations of acoustic feature extractors to be used
18
19
  data_name (string): names of databases that are extracted (for the caching)
19
- feats_designation (string): the type of split (train/test), also is used for the cache name.
20
+ feats_designation (string): the type of split (train/test), also is used for the cache name.
20
21
  Returns:
21
- df (pandas.DataFrame): dataframe with same index as data_df and acoustic features in columns
22
+ df (pandas.DataFrame): dataframe with same index as data_df and acoustic features in columns
22
23
  """
23
- df = None # pandas dataframe to store the features (and indexed with the data from the sets)
24
- data_df = None # dataframe to get audio paths
25
24
 
26
- # def __init__
25
+ # pandas dataframe to store the features (and indexed with the data from the sets)
26
+ df = None
27
+ data_df = None # dataframe to get audio paths
28
+
29
+ # def __init__
27
30
  def __init__(self, data_df, feats_types, data_name, feats_designation):
28
31
  self.data_df = data_df
29
32
  self.data_name = data_name
30
33
  self.feats_types = feats_types
31
- self.util = Util('feature_extractor')
34
+ self.util = Util("feature_extractor")
32
35
  self.feats_designation = feats_designation
33
-
36
+
34
37
  def extract(self):
35
38
  # feats_types = self.util.config_val_list('FEATS', 'type', ['os'])
36
39
  self.featExtractor = None
37
- self.feats= pd.DataFrame()
40
+ self.feats = pd.DataFrame()
38
41
  _scale = True
39
42
  for feats_type in self.feats_types:
40
- store_name = f'{self.data_name}_{feats_type}'
41
- if feats_type=='os':
42
- self.featExtractor = Opensmileset(f'{store_name}_{self.feats_designation}', self.data_df)
43
- elif feats_type=='trill':
43
+ store_name = f"{self.data_name}_{feats_type}"
44
+ if feats_type == "os":
45
+ self.featExtractor = Opensmileset(
46
+ f"{store_name}_{self.feats_designation}", self.data_df
47
+ )
48
+ elif feats_type == "trill":
44
49
  from nkululeko.feat_extract.feats_trill import TRILLset
45
- self.featExtractor = TRILLset(f'{store_name}_{self.feats_designation}', self.data_df)
46
- elif feats_type=='wav2vec':
50
+ self.featExtractor = TRILLset(
51
+ f"{store_name}_{self.feats_designation}", self.data_df
52
+ )
53
+ elif feats_type == "wav2vec":
47
54
  from nkululeko.feat_extract.feats_wav2vec2 import Wav2vec2
48
- self.featExtractor = Wav2vec2(f'{store_name}_{self.feats_designation}', self.data_df)
49
- elif feats_type=='audmodel':
55
+ self.featExtractor = Wav2vec2(
56
+ f"{store_name}_{self.feats_designation}", self.data_df
57
+ )
58
+ elif feats_type in ("hubert", "hubert_ft", "hubert_large",
59
+ "hubert_xlarge", "hubert_xlarge_ft"):
60
+ from nkululeko.feat_extract.feats_hubert import Hubert
61
+ self.featExtractor = Hubert(
62
+ f"{store_name}_{self.feats_designation}", self.data_df,
63
+ feats_type
64
+ )
65
+
66
+ elif feats_type == "audmodel":
50
67
  from nkululeko.feat_extract.feats_audmodel import AudModelSet
51
- self.featExtractor = AudModelSet(f'{store_name}_{self.feats_designation}', self.data_df)
52
- elif feats_type=='auddim':
53
- from nkululeko.feat_extract.feats_audmodel_dim import AudModelDimSet
54
- self.featExtractor = AudModelDimSet(f'{store_name}_{self.feats_designation}', self.data_df)
55
- elif feats_type=='agender':
56
- from nkululeko.feat_extract.feats_agender import AudModelAgenderSet
57
- self.featExtractor = AudModelAgenderSet(f'{store_name}_{self.feats_designation}', self.data_df)
58
- elif feats_type=='agender_agender':
59
- from nkululeko.feat_extract.feats_agender_agender import AgenderAgenderSet
60
- self.featExtractor = AgenderAgenderSet(f'{store_name}_{self.feats_designation}', self.data_df)
61
- elif feats_type=='snr':
68
+ self.featExtractor = AudModelSet(
69
+ f"{store_name}_{self.feats_designation}", self.data_df
70
+ )
71
+ elif feats_type == "auddim":
72
+ from nkululeko.feat_extract.feats_audmodel_dim import \
73
+ AudModelDimSet
74
+ self.featExtractor = AudModelDimSet(
75
+ f"{store_name}_{self.feats_designation}", self.data_df
76
+ )
77
+ elif feats_type == "agender":
78
+ from nkululeko.feat_extract.feats_agender import \
79
+ AudModelAgenderSet
80
+ self.featExtractor = AudModelAgenderSet(
81
+ f"{store_name}_{self.feats_designation}", self.data_df
82
+ )
83
+ elif feats_type == "agender_agender":
84
+ from nkululeko.feat_extract.feats_agender_agender import \
85
+ AgenderAgenderSet
86
+ self.featExtractor = AgenderAgenderSet(
87
+ f"{store_name}_{self.feats_designation}", self.data_df
88
+ )
89
+ elif feats_type == "snr":
62
90
  from nkululeko.feat_extract.feats_snr import SNRSet
63
- self.featExtractor = SNRSet(f'{store_name}_{self.feats_designation}', self.data_df)
64
- elif feats_type=='mos':
91
+ self.featExtractor = SNRSet(
92
+ f"{store_name}_{self.feats_designation}", self.data_df
93
+ )
94
+ elif feats_type == "mos":
65
95
  from nkululeko.feat_extract.feats_mos import MOSSet
66
- self.featExtractor = MOSSet(f'{store_name}_{self.feats_designation}', self.data_df)
67
- elif feats_type=='squim':
96
+ self.featExtractor = MOSSet(
97
+ f"{store_name}_{self.feats_designation}", self.data_df
98
+ )
99
+ elif feats_type == "squim":
68
100
  from nkululeko.feat_extract.feats_squim import SQUIMSet
69
- self.featExtractor = SQUIMSet(f'{store_name}_{self.feats_designation}', self.data_df)
70
- elif feats_type=='clap':
101
+ self.featExtractor = SQUIMSet(
102
+ f"{store_name}_{self.feats_designation}", self.data_df
103
+ )
104
+ elif feats_type == "clap":
71
105
  from nkululeko.feat_extract.feats_clap import Clap
72
- self.featExtractor = Clap(f'{store_name}_{self.feats_designation}', self.data_df)
73
- elif feats_type=='praat':
106
+ self.featExtractor = Clap(
107
+ f"{store_name}_{self.feats_designation}", self.data_df
108
+ )
109
+ elif feats_type == "praat":
74
110
  from nkululeko.feat_extract.feats_praat import Praatset
75
- self.featExtractor = Praatset(f'{store_name}_{self.feats_designation}', self.data_df)
76
- elif feats_type=='mld':
111
+ self.featExtractor = Praatset(
112
+ f"{store_name}_{self.feats_designation}", self.data_df
113
+ )
114
+ elif feats_type == "mld":
77
115
  from nkululeko.feat_extract.feats_mld import MLD_set
78
- self.featExtractor = MLD_set(f'{store_name}_{self.feats_designation}', self.data_df)
79
- elif feats_type=='import':
116
+ self.featExtractor = MLD_set(
117
+ f"{store_name}_{self.feats_designation}", self.data_df
118
+ )
119
+ elif feats_type == "import":
80
120
  from nkululeko.feat_extract.feats_import import Importset
81
- self.featExtractor = Importset(f'{store_name}_{self.feats_designation}', self.data_df)
121
+ self.featExtractor = Importset(
122
+ f"{store_name}_{self.feats_designation}", self.data_df
123
+ )
82
124
  else:
83
- self.util.error(f'unknown feats_type: {feats_type}')
125
+ self.util.error(f"unknown feats_type: {feats_type}")
84
126
 
85
127
  self.featExtractor.extract()
86
128
  self.featExtractor.filter()
87
129
  # remove samples that were not extracted by MLD
88
- #self.df_test = self.df_test.loc[self.df_test.index.intersection(featExtractor_test.df.index)]
89
- #self.df_train = self.df_train.loc[self.df_train.index.intersection(featExtractor_train.df.index)]
90
- self.util.debug(f'{feats_type}: shape : {self.featExtractor.df.shape}')
91
- self.feats = pd.concat([self.feats, self.featExtractor.df], axis = 1)
130
+ # self.df_test = self.df_test.loc[self.df_test.index.intersection(featExtractor_test.df.index)]
131
+ # self.df_train = self.df_train.loc[self.df_train.index.intersection(featExtractor_train.df.index)]
132
+ self.util.debug(
133
+ f"{feats_type}: shape : {self.featExtractor.df.shape}")
134
+ self.feats = pd.concat([self.feats, self.featExtractor.df], axis=1)
92
135
  return self.feats
93
136
 
94
137
  def extract_sample(self, signal, sr):
95
- return self.featExtractor.extract_sample(signal, sr)
138
+ return self.featExtractor.extract_sample(signal, sr)
nkululeko/plots.py CHANGED
@@ -95,6 +95,17 @@ class Plots():
95
95
  else:
96
96
  self.util.error(f'plot value counts: the plot distribution descriptor for {att} has more than 2 values')
97
97
 
98
+ def plot_durations(self, df, filename, sample_selection):
99
+ fig_dir = self.util.get_path('fig_dir')+'../' # one up because of the runs
100
+ ax = sns.histplot(df, x='duration', hue='class_label', kde=True)
101
+ ax.set_title(f'{sample_selection} {df.shape[0]}')
102
+ ax.set_xlabel(f'duration')
103
+ ax.set_ylabel(f'number of samples')
104
+ fig = ax.figure
105
+ plt.tight_layout()
106
+ plt.savefig(f'{fig_dir}{filename}_{sample_selection}.{self.format}')
107
+ plt.close(fig)
108
+ fig.clear()
98
109
 
99
110
  def describe_df(self, name, df, target, filename):
100
111
  """Make a stacked barplot of samples and speakers per sex and target values. speaker, gender and target columns must be present"""
nkululeko/segment.py CHANGED
@@ -7,25 +7,8 @@ from nkululeko.util import Util
7
7
  from nkululeko.constants import VERSION
8
8
  import argparse
9
9
  import os
10
- import torch
11
- import audformat
12
- from audformat.utils import to_filewise_index
13
- from audformat import segmented_index
14
10
  import pandas as pd
15
11
 
16
- # initialize the VAD model
17
- SAMPLING_RATE = 16000
18
- torch.set_num_threads(1)
19
- vad_model, vad_utils = torch.hub.load(repo_or_dir='snakers4/silero-vad',
20
- model='silero_vad',
21
- force_reload=False,
22
- onnx=False)
23
- (get_speech_timestamps,
24
- save_audio,
25
- read_audio,
26
- VADIterator,
27
- collect_chunks) = vad_utils
28
-
29
12
  def main(src_dir):
30
13
  parser = argparse.ArgumentParser(description='Call the nkululeko framework.')
31
14
  parser.add_argument('--config', default='exp.ini', help='The base configuration')
@@ -60,7 +43,7 @@ def main(src_dir):
60
43
  util.debug(f'train shape : {expr.df_train.shape}, test shape:{expr.df_test.shape}')
61
44
 
62
45
  # segment
63
- segment_target = util.config_val('DATA', 'segment_target', '_seg')
46
+ segment_target = util.config_val('SEGMENT', 'target', '_seg')
64
47
  # this if a specific dataset is to be segmented
65
48
  # segment_db = util.config_val('DATA', 'segment', False)
66
49
  # if segment_db:
@@ -72,7 +55,8 @@ def main(src_dir):
72
55
  # name = f'{dataset}{segment_target}'
73
56
  # df_seg.to_csv(f'{expr.data_dir}/{name}.csv')
74
57
 
75
- sample_selection = util.config_val('DATA', 'segment', 'all')
58
+ segmenter = util.config_val('SEGMENT', 'method', 'silero')
59
+ sample_selection = util.config_val('SEGMENT', 'sample_selection', 'all')
76
60
  if sample_selection=='all':
77
61
  df = pd.concat([expr.df_train, expr.df_test])
78
62
  elif sample_selection=='train':
@@ -81,26 +65,38 @@ def main(src_dir):
81
65
  df = expr.df_test
82
66
  else:
83
67
  util.error(f'unknown segmentation selection specifier {sample_selection}, should be [all | train | test]')
68
+ # if "duration" not in df.columns:
69
+ # df = df.drop(columns=['duration'], inplace=True)
70
+ util.debug(f'segmenting {sample_selection}: {df.shape[0]} samples with {segmenter}')
71
+ if segmenter=='silero':
72
+ from nkululeko.segmenting.seg_silero import Silero_segmenter
73
+ segmenter = Silero_segmenter()
74
+ df_seg = segmenter.segment_dataframe(df)
75
+
76
+ else:
77
+ util.error(f'unkown segmenter: {segmenter}')
84
78
 
85
- if "duration" not in df.columns:
86
- df = df.drop(columns=['duration'], inplace=True)
87
- util.debug(f'segmenting train and test set: {df.shape[0]} samples')
88
- df_seg = segment_dataframe(df)
89
79
  def calc_dur(x):
80
+ from datetime import datetime
90
81
  starts = x[1]
91
82
  ends = x[2]
92
83
  return (ends - starts).total_seconds()
93
84
  df_seg['duration'] = df_seg.index.to_series().map(lambda x:calc_dur(x))
85
+ if "duration" not in df.columns:
86
+ df['duration'] = df.index.to_series().map(lambda x:calc_dur(x))
87
+ num_before = df.shape[0]
88
+ num_after = df_seg.shape[0]
94
89
  dataname = '_'.join(expr.datasets.keys())
95
90
  name = f'{dataname}{segment_target}'
96
91
  df_seg.to_csv(f'{expr.data_dir}/{name}.csv')
92
+ from nkululeko.plots import Plots
93
+ plots = Plots()
94
+ plots.plot_durations(df, 'original_durations', sample_selection)
95
+ plots.plot_durations(df_seg, 'segmented_durations', sample_selection)
97
96
  print('')
98
- util.debug(f'saved {name}.csv to {expr.data_dir}, {df_seg.shape[0]} samples')
97
+ util.debug(f'saved {name}.csv to {expr.data_dir}, {num_after} samples (was {num_before})')
99
98
  print('DONE')
100
99
 
101
-
102
-
103
-
104
100
  def get_segmentation(file):
105
101
  # print(f'segmenting {file[0]}')
106
102
  print('.', end='')
{nkululeko-0.60.0.dist-info → nkululeko-0.61.0.dist-info}/METADATA RENAMED
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.1
2
2
  Name: nkululeko
3
- Version: 0.60.0
3
+ Version: 0.61.0
4
4
  Summary: Machine learning audio prediction experiments based on templates
5
5
  Home-page: https://github.com/felixbur/nkululeko
6
6
  Author: Felix Burkhardt
@@ -253,6 +253,10 @@ Nkululeko can be used under the [MIT license](https://choosealicense.com/license
253
253
  Changelog
254
254
  =========
255
255
 
256
+ Version 0.61.0
257
+ --------------
258
+ * added HUBERT embeddings
259
+
256
260
  Version 0.60.0
257
261
  --------------
258
262
  * some bugfixes
{nkululeko-0.60.0.dist-info → nkululeko-0.61.0.dist-info}/RECORD RENAMED
@@ -2,30 +2,30 @@ nkululeko/__init__.py,sha256=62f8HiEzJ8rG2QlTFJXUCMpvuH3fKI33DoJSj33mscc,63
2
2
  nkululeko/augment.py,sha256=ebv5QebGD8wLzXInvusjn4kFlET6-yXkYoF132BrubQ,1750
3
3
  nkululeko/balancer.py,sha256=64ftZN68sMDfkvuovCDHpAHmSJgCO6Kdk9bwmpSisec,12
4
4
  nkululeko/cacheddataset.py,sha256=bSJ_SDg7TxL89YL_pJXp-sFvdUXJtHuBTd5KSTE4AkQ,955
5
- nkululeko/constants.py,sha256=bmZ-IYOBLr7300ZA6Lk76zK1bBaZbY-EMByZnpKFkiI,18
5
+ nkululeko/constants.py,sha256=xSkfM3CWCYGD40D6GhHHCrxrw0JNK-372YdiRKMknWs,18
6
6
  nkululeko/demo.py,sha256=nGP3fUDXuW1ZF12AzMpzRWXct0rdqYRJVNgA9B_QWwA,1821
7
7
  nkululeko/demo_predictor.py,sha256=VVxE2lf5lTkAP5qElG5U2bK6SdDzQ2Jmf0Vn_yHpSro,2302
8
8
  nkululeko/experiment.py,sha256=9qStgy31svY4bBVZOkuJ0JFjEQ1sIT2ibIdJ6IVlfTI,25063
9
9
  nkululeko/explore.py,sha256=aemOk5XYw7axQEJQfdABEUxN3img0NYSb8l6W-nDzZY,2090
10
- nkululeko/feature_extractor.py,sha256=ZNNgf5u_daxoJ7Ys4jtT4tYD2PNNyV950Pa_R1c2gPM,5297
10
+ nkululeko/feature_extractor.py,sha256=2LqPIiDAoaBRhjcKik2hjBEBVBsLLxx8blQvTD43TRg,6324
11
11
  nkululeko/file_checker.py,sha256=Nw05SIp7Ez1U9ZeFhNGz0XivwKr43hHg1WsfzKsrFPQ,3510
12
12
  nkululeko/filter_data.py,sha256=g7giEShbA-dr2ekVycW5WurFG-UaopJvDZWylKNZtpM,6717
13
13
  nkululeko/glob_conf.py,sha256=2Tl0NZQeVeydDO8T2tuJC2lCv99wvaTVRX9Dy1Mq_L4,237
14
14
  nkululeko/modelrunner.py,sha256=zVDi2-UyjtmU0_Ltf4lnPcECVtukuDVuZaj4pydqOBY,5478
15
15
  nkululeko/nkululeko.py,sha256=O2Zw7u-Mb7VP9MPxAlhdTkXV2lW2kETIuSJp7mfj_Tc,1671
16
- nkululeko/plots.py,sha256=rr4Fa0kckRNw5-mz9clikiB5ul8XJ9U4_1OebCUiNWw,10924
16
+ nkululeko/plots.py,sha256=hoOFLbWXpV5jGDWHEpy345_4vpaGKGMAv2JwvpNUxkw,11454
17
17
  nkululeko/predict.py,sha256=3ei4wn2by0p9Vkv7cllMcszmEjSM2vX0T6x_5rlgT28,1851
18
18
  nkululeko/reporter.py,sha256=359aeQWt0ZGLseaJnOfafYG8BrwumiM2Q58DWiaoyWQ,10177
19
19
  nkululeko/result.py,sha256=Ey5FPsAyfnQVtzO_J6_4hkOAZ191YWmF_vXxlgNjCdc,406
20
20
  nkululeko/runmanager.py,sha256=ll04dEu5Y1nOi8QOtmSiw3oxzcXeASdQsg0t-vxCny8,6765
21
21
  nkululeko/scaler.py,sha256=6NQHbSQZO9HIfhYNlliuDRywjaEH_FVKHRskTJ2xr90,3021
22
- nkululeko/segment.py,sha256=xmkZrA3iEsro7vaBwWdey_pIleS1tYyjTFe-EfMUQHQ,4399
22
+ nkululeko/segment.py,sha256=GGyovnZ75Sqd8TgBH5fi3fjRkVw_ygqBQD46Yn6GVQ4,4660
23
23
  nkululeko/syllable_nuclei.py,sha256=vK9dj5deqRnyEmlZmhFtKPzqKVGNCgTnWaG8UDITKNg,9913
24
24
  nkululeko/test.py,sha256=BbHGliDChAXqMe2oA579dJpyZSlPGAm5997lX_POboQ,1372
25
25
  nkululeko/test_predictor.py,sha256=QwdAVPHNew9w5PD_sPFhhWVDTYRAbUE6fkAp58X8Hjg,2410
26
26
  nkululeko/util.py,sha256=gZrNTF4C7hKkEMCC_hoNkEAhAViWzWebP8LsHRew7s4,9731
27
- nkululeko-0.60.0.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
28
- nkululeko-0.60.0.dist-info/METADATA,sha256=p1kHUf0egmZQHlH4RbGg2MdbYf9GmMvuW37LHMNWmvY,21621
29
- nkululeko-0.60.0.dist-info/WHEEL,sha256=yQN5g4mg4AybRjkgi-9yy4iQEFibGQmlz78Pik5Or-A,92
30
- nkululeko-0.60.0.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
31
- nkululeko-0.60.0.dist-info/RECORD,,
27
+ nkululeko-0.61.0.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
28
+ nkululeko-0.61.0.dist-info/METADATA,sha256=LXJjW9KpGkPum60eGuYW__gl5QKXVRhnm6RySrKo2b8,21680
29
+ nkululeko-0.61.0.dist-info/WHEEL,sha256=yQN5g4mg4AybRjkgi-9yy4iQEFibGQmlz78Pik5Or-A,92
30
+ nkululeko-0.61.0.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
31
+ nkululeko-0.61.0.dist-info/RECORD,,