It uses the Wavelet Covariance Matrix between pairs of audio recordings as a measure of similarity for Blind Speaker Verification. Authors: Angie Natalia Vasquez, Dora Maria Ballesteros, and Diego Renza.
1. Use data Speakers.csv 2. Use module Clean Missing Data to replace the Nan or Inf values by the median of the data. 3. Use module Split Data: 60% for training; 40% for testing. 4. Use module Two-Class Neural Network. 5. Use module Train Model where the output column named as Output. The inputs are Two-Class Neural Network and Split Data (the left part). The output is Score Model. 6. For Score Model the inputs are Train Model and Split Data (the right part). 7. Use Evaluate Model to obtain the performance. The performance results with the validation dataset are: Accuracy of 88.2%, Precision of 84.5%, Recall of 90%, F1 of 87.2% and an AUC of 93.8% For detail information: Speaker Verification with fake intonation based on Neural Networks. 7th IAPR/IEEE International Workshop on Biometrics and Forensic - IWBF 2019.