nkululeko 0.48.1__tar.gz → 0.49.0__tar.gz

{nkululeko-0.48.1 → nkululeko-0.49.0}/CHANGELOG.md

@@ -1,6 +1,10 @@
 Changelog
 =========
 
+Version 0.49.0
+--------------
+* added random-splicing
+
 Version 0.48.1
 --------------
 * bugfix: database object was not loaded when dataframe was reused

{nkululeko-0.48.1/nkululeko.egg-info → nkululeko-0.49.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nkululeko
-Version: 0.48.1
+Version: 0.49.0
 Summary: Machine learning audio prediction experiments based on templates
 Home-page: https://github.com/felixbur/nkululeko
 Author: Felix Burkhardt
@@ -214,6 +214,10 @@ Nkululeko can be used under the [MIT license](https://choosealicense.com/license
 Changelog
 =========
 
+Version 0.49.0
+--------------
+* added random-splicing
+
 Version 0.48.1
 --------------
 * bugfix: database object was not loaded when dataframe was reused

{nkululeko-0.48.1 → nkululeko-0.49.0}/nkululeko/augment.py

@@ -38,7 +38,13 @@ def main(src_dir):
     util.debug(f'train shape : {expr.df_train.shape}, test shape:{expr.df_test.shape}')
 
     # augment
-    expr.augment()
+    augmenting = util.config_val('DATA', 'augment', False)
+    if augmenting:
+        expr.augment()
+    
+    random_splicing = util.config_val('DATA', 'random_splice', False)
+    if random_splicing:
+        expr.random_splice()
 
     print('DONE')
 

{nkululeko-0.48.1 → nkululeko-0.49.0}/nkululeko/augmenter.py

@@ -12,8 +12,8 @@ class Augmenter:
     """
         augmenting the train split
     """
-    def __init__(self, train_df):
-        self.train_df = train_df
+    def __init__(self, df):
+        self.df = df
         self.util = Util()
         # Define a standard transformation that randomly add augmentations to files
         self.audioment = Compose([
@@ -30,15 +30,15 @@ class Augmenter:
  #       audeer.mkdir(newpath)
         return fp.replace(fullpath, np)
 
-    def augment(self):
+    def augment(self, sample_selection):
         """
         augment the training files and return a dataframe with new files index.
         """
-        files = self.train_df.index.get_level_values(0).values
+        files = self.df.index.get_level_values(0).values
         store = self.util.get_path('store')
         filepath = f'{store}augmentations/'
         audeer.mkdir(filepath)
-        self.util.debug(f'augmenting the training set to {filepath}')
+        self.util.debug(f'augmenting {sample_selection} samples to {filepath}')
         newpath = ''
         for i, f in enumerate(files):
             signal, sr = audiofile.read(f)
@@ -50,9 +50,9 @@ class Augmenter:
             audiofile.write(f'{newpath}{filename}', signal=sig_aug, sampling_rate=sr)
             if i%10==0:
                 print(f'augmented {i} of {len(files)}')
-        df_ret = self.train_df.copy()
+        df_ret = self.df.copy()
         df_ret = df_ret.set_index(map_file_path(df_ret.index, lambda x: self.changepath(x, newpath)))
-        aug_db_filename = self.util.config_val('DATA', 'augment', 'augment.csv')
+        aug_db_filename = self.util.config_val('DATA', 'augment_result', 'augment.csv')
         target = self.util.config_val('DATA', 'target', 'emotion')
         df_ret[target] = df_ret['class_label']
         df_ret = df_ret.drop(columns=['class_label'])

nkululeko-0.49.0/nkululeko/constants.py

@@ -0,0 +1 @@
+VERSION = '0.49.0'

{nkululeko-0.48.1 → nkululeko-0.49.0}/nkululeko/experiment.py

@@ -292,18 +292,38 @@ class Experiment:
 
     def augment(self):
         """
-        Augment the training set
+        Augment the selected samples
         """
         from nkululeko.augmenter import Augmenter
-        augmenter = Augmenter(self.df_train)
-        augmenter.augment()
-
-    # def augment_train(self):
-    #     """Augment the train dataframe"""
-    #     from nkululeko.augmenter import Augmenter
-    #     augment_train = Augmenter(self.df_train)
-    #     df_train_aug = augment_train.augment()
-    #     self.df_train = self.df_train.append(df_train_aug)
+        sample_selection = self.util.config_val(' DATA', 'augment', 'train')
+        if sample_selection=='all':
+            df = pd.concat([self.df_train, self.df_test])
+        elif sample_selection=='train':
+            df = self.df_train
+        elif sample_selection=='test':
+            df = self.df_test
+        else:
+            self.util.error(f'unknown augmentation selection specifier {sample_selection}, should be [all | train | test]')
+
+        augmenter = Augmenter(df)
+        augmenter.augment(sample_selection)
+
+    def random_splice(self):
+        """
+        Random-splice the selected samples
+        """
+        from nkululeko.randomsplicer import Randomsplicer
+        sample_selection = self.util.config_val(' DATA', 'random_splice', 'train')
+        if sample_selection=='all':
+            df = pd.concat([self.df_train, self.df_test])
+        elif sample_selection=='train':
+            df = self.df_train
+        elif sample_selection=='test':
+            df = self.df_test
+        else:
+            self.util.error(f'unknown augmentation selection specifier {sample_selection}, should be [all | train | test]')
+        randomsplicer = Randomsplicer(df)
+        randomsplicer.run(sample_selection)
 
 
     def analyse_features(self, needs_feats):

{nkululeko-0.48.1 → nkululeko-0.49.0}/nkululeko/feinberg_praat.py

@@ -119,7 +119,6 @@ def runPCA(df):
     # PCA
     pca = PCA(n_components=2)
     principalComponents = pca.fit_transform(x)
-    print(type(principalComponents))
     if np.any(np.isnan(principalComponents)):
         print ('pc is nan')
         print(f'count: {np.count_nonzero(np.isnan(principalComponents))}')

nkululeko-0.49.0/nkululeko/randomsplicer.py

@@ -0,0 +1,76 @@
+# randomsplicer.py
+
+"""
+Code originally by Oliver Pauly
+
+Based on an idea by Klaus Scherer
+
+K. R. Scherer, “Randomized splicing: A note on a simple technique for masking speech content” 
+Journal of Experimental Research in Personality, vol. 5, pp. 155–159, 1971.
+
+Evaluated in:
+F. Burkhardt, Anna Derington, Matthias Kahlau, Klaus Scherer, Florian Eyben and Björn Schuller: Masking Speech Contents by Random Splicing: is Emotional Expression Preserved?, Proc. ICASSP, 2023
+
+"""
+
+import pandas as pd
+from nkululeko.util import Util 
+import nkululeko.randomsplicing as rsp
+import numpy as np
+import audiofile as af
+import os
+from audformat.utils import map_file_path
+import audeer
+
+class Randomsplicer:
+    """
+        augmenting the train split
+    """
+    def __init__(self, df):
+        self.df = df
+        self.util = Util()
+
+    def changepath(self, fp, np):
+        fullpath = os.path.dirname(fp)
+        return fp.replace(fullpath, np)
+
+    def run(self, sample_selection):
+        """
+        random splice the selected samples and return a dataframe with new files index.
+            adjustable parameters:
+            * p_reverse: probability of some samples to be in reverse order (default: 0.3)
+            * top_db: top db level for silence to be recognized (default: 12)   
+        """
+
+        p_reverse=0.3
+        top_db=12
+
+        files = self.df.index.get_level_values(0).values
+        store = self.util.get_path('store')
+        filepath = f'{store}randomspliced/'
+        audeer.mkdir(filepath)
+        self.util.debug(f'random splicing {sample_selection} samples to {filepath}')
+        newpath = ''
+        for i, f in enumerate(files):
+            signal, sr = af.read(f)
+            filename = os.path.basename(f)
+            parent = os.path.dirname(f).split('/')[-1]
+            sig_new = rsp.random_splicing(
+                signal, sr,
+                p_reverse=p_reverse,
+                top_db=top_db,
+                )
+
+            newpath = f'{filepath}/{parent}/'
+            audeer.mkdir(newpath)
+            af.write(f'{newpath}{filename}', signal=sig_new, sampling_rate=sr)
+            if i%10==0:
+                print(f'random spliced {i} of {len(files)}')
+        df_ret = self.df.copy()
+        df_ret = df_ret.set_index(map_file_path(df_ret.index, lambda x: self.changepath(x, newpath)))
+        db_filename = self.util.config_val('DATA', 'random_splice_result', 'random_spliced.csv')
+        target = self.util.config_val('DATA', 'target', 'emotion')
+        df_ret[target] = df_ret['class_label']
+        df_ret = df_ret.drop(columns=['class_label'])
+        df_ret.to_csv(db_filename)
+        return df_ret

nkululeko-0.49.0/nkululeko/randomsplicing.py

@@ -0,0 +1,74 @@
+"""
+Code originally by Oliver Pauly
+
+Based on an idea by Klaus Scherer
+
+K. R. Scherer, “Randomized splicing: A note on a simple technique for masking speech content” 
+Journal of Experimental Research in Personality, vol. 5, pp. 155–159, 1971.
+
+Evaluated in:
+F. Burkhardt, Anna Derington, Matthias Kahlau, Klaus Scherer, Florian Eyben and Björn Schuller: Masking Speech Contents by Random Splicing: is Emotional Expression Preserved?, Proc. ICASSP, 2023
+
+"""
+import numpy as np
+import librosa
+import audiofile as af
+
+def random_splicing(
+    signal, sr,
+    p_reverse=0.0,
+    top_db=12,
+    ):
+    """
+    randomly splice the signal and re-arrange.    
+    
+    p_reverse: probability of some samples to be in reverse order
+    top_db:  top db level for silence to be recognized
+    
+    """
+    signal /= np.max(abs(signal))
+
+    indices = split_wav_naive(signal, top_db=top_db)
+
+    np.random.shuffle(indices) 
+
+    wav_spliced = remix_random_reverse(signal, indices, p_reverse=p_reverse)
+        
+    return wav_spliced
+
+
+def split_wav_naive(wav, top_db=12):
+      
+    indices = librosa.effects.split(wav, top_db=top_db)
+    
+    indices = np.array(indices)
+    # (re)add the silence-segments
+    indices = indices.repeat(2)[1:-1].reshape((-1, 2))
+    # add first segment
+    indices = np.vstack(((0, indices[0][0]), indices))
+    # add last segment
+    indices = np.vstack((indices, [indices[-1][-1], wav.shape[0]]))
+
+    return indices
+
+
+def remix_random_reverse(wav, indices, p_reverse=0):
+    
+    import random
+
+    wav_remix = []
+    
+    for seg in indices:
+        
+        start = seg[0]
+        end = seg[1]
+        wav_seg = wav[start:end]
+            
+        if np.random.rand(1)[0] <= p_reverse:
+            wav_seg = wav_seg[::-1]
+    
+        wav_remix.append(wav_seg)
+        
+    wav_remix = np.hstack(wav_remix)
+    
+    return wav_remix

{nkululeko-0.48.1 → nkululeko-0.49.0/nkululeko.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nkululeko
-Version: 0.48.1
+Version: 0.49.0
 Summary: Machine learning audio prediction experiments based on templates
 Home-page: https://github.com/felixbur/nkululeko
 Author: Felix Burkhardt
@@ -214,6 +214,10 @@ Nkululeko can be used under the [MIT license](https://choosealicense.com/license
 Changelog
 =========
 
+Version 0.49.0
+--------------
+* added random-splicing
+
 Version 0.48.1
 --------------
 * bugfix: database object was not loaded when dataframe was reused

{nkululeko-0.48.1 → nkululeko-0.49.0}/nkululeko.egg-info/SOURCES.txt

@@ -52,6 +52,8 @@ nkululeko/model_xgr.py
 nkululeko/modelrunner.py
 nkululeko/nkululeko.py
 nkululeko/plots.py
+nkululeko/randomsplicer.py
+nkululeko/randomsplicing.py
 nkululeko/reporter.py
 nkululeko/result.py
 nkululeko/runmanager.py

nkululeko-0.48.1/nkululeko/constants.py

@@ -1 +0,0 @@
-VERSION = '0.48.1'