Regarding the dataset on which the model is trained

I am using actor speech data which contains around 1500 audio files. But whenever i am extracting features from the data, I am getting only 990 rows. And model accuracy is around 46%

Is there any other source data from where the data has been trained? 