Fixed issues in ModelsTrainer and SpeakerIdentifier

Code/ModelsTrainer.py

@@ -2,9 +2,9 @@
 import pickle
 import warnings
 import numpy as np
-from sklearn.mixture import GMM
+from sklearn.mixture import GaussianMixture as GMM
 from FeaturesExtractor import FeaturesExtractor
-from SilenceEliminator import SilenceEliminator
+import scipy.io.wavfile
 
 warnings.filterwarnings("ignore")
 
@@ -34,12 +34,10 @@
 
         # extract voice features
         features_extractor = FeaturesExtractor()
-        silence_eliminator = SilenceEliminator()
 
         try   :
-            silence_eliminated_wave_file_path = "temp-" + os.path.basename(filepath).split('.')[0] + ".wav"
-            audio, duration_string = silence_eliminator.ffmpeg_silence_eliminator(filepath, silence_eliminated_wave_file_path)
-            vector                 = features_extractor.accelerated_get_features_vector(filepath, audio, 8000)
+            sample_rate, signal = scipy.io.wavfile.read(filepath)
+            vector                 = features_extractor.accelerated_get_features_vector(filepath, signal, 8000)
         except:
             continue
 
@@ -52,7 +50,7 @@
                 print("ValueError: Shape mismatch")
 
     # adapt gmm
-    gmm = GMM(n_components = 16, n_iter = 200, covariance_type='diag', n_init = 3)
+    gmm = GMM(n_components = 16, covariance_type='diag', n_init = 3)
     gmm.fit(features)
 
     # dumping the trained gaussian model

Code/SpeakerIdentifier.py

@@ -2,8 +2,8 @@
 import pickle
 import warnings
 import numpy as np
-from SilenceEliminator import SilenceEliminator
 from FeaturesExtractor import FeaturesExtractor
+import scipy.io.wavfile
 
 warnings.filterwarnings("ignore")
 
@@ -32,11 +32,9 @@
 for path in file_paths[:]:
     if os.path.basename(path).split('_')[0] in db.keys():
         features_extractor = FeaturesExtractor()
-        silence_eliminator = SilenceEliminator()
 
-        silence_eliminated_wave_file_path ="temp-" + os.path.basename(path).split('.')[0] + ".wav"
-        audio, duration_string = silence_eliminator.ffmpeg_silence_eliminator(path, silence_eliminated_wave_file_path)
-        vector                 = features_extractor.accelerated_get_features_vector(path, audio, 8000)
+        sample_rate, signal = scipy.io.wavfile.read(path)
+        vector                 = features_extractor.accelerated_get_features_vector(path, signal, 8000)
 
         if vector.shape != (0,):
             print(vector.shape)

README.md

@@ -42,3 +42,9 @@ This script require the follwing modules/libraries:
 ## Results and disscussion
 - The code can be further optimized using multi-threading, acceleration libs and multi-processing.
 - The accuracy can be further improved using GMM normalization aka a UBM-GMM system.
+
+## Update
+- Eliminated the use of SilenceEliminator
+- Achieved an accuracy of 98.21%
+- Fixed the import of GMM in FeaturesExtractor
+- Updated file_read in ModelsTrainer

Run.py

@@ -4,19 +4,18 @@
 
 
 if __name__== "__main__":
-   # download dataset
-   print("# Download dataset zip file")
-   zip_url  = "http://www.openslr.org/resources/45/ST-AEDS-20180100_1-OS.tgz"
-   urllib.request.urlretrieve(zip_url, 'SLR45.tgz')
-
-   # extract and manage dataset files
-   print("# Mange and organize files")
-   os.system('python3 Code/DataManager.py')
-
-   # train speakers gmm models
-   print("# Train gender models")
-   os.system('python3 Code/ModelsTrainer.py')
-
-    # test system and recognise/identify speakers
-    print(" # Identify genders")
-    os.system('python3 Code/SpeakerIdentifier.py')
+	# download dataset
+	print("# Download dataset zip file")
+	zip_url  = "http://www.openslr.org/resources/45/ST-AEDS-20180100_1-OS.tgz"
+	urllib.request.urlretrieve(zip_url, 'SLR45.tgz')
+
+	# extract and manage dataset files
+	print("# Mange and organize files")
+	os.system('python3 Code/DataManager.py')
+	# train speakers gmm models
+	print("# Train gender models")
+	os.system('python3 Code/ModelsTrainer.py')
+
+	# test system and recognise/identify speakers
+	print(" # Identify genders")
+	os.system('python3 Code/SpeakerIdentifier.py')