Comparative study of CNN, LSTM and hybrid CNN-LSTM model in amazigh speech recognition using spectrogram feature extraction and different gender and age dataset