0
$\begingroup$

I am working on a deepfake audio classification project with a dataset consisting of only 3000 samples. I have made several attempts to address this challenge. Firstly, I extracted melspectrograms and utilized them as input for CNN and BiLSTM models. Also, I extracted various features such as chroma_stft, rms, spectral_centroid, rolloff, zero_crossing_rate, and mfcc, and passed them in different machine learning models including SVM, RF, XGBoost, and others.However, I have concerns regarding the generalizability of my model. I am unsure if it will perform well on unseen data. Hence, I would greatly appreciate any suggestions for alternative approaches or recommendations for pretrained models that might be suitable for this particular case.

$\endgroup$

0

You must log in to answer this question.