There's like only 40 seconds of him actually talking

Also how long is the dataset for this? And did you use lossless audio? I feel like you could train this model more given that there's a lot of audio of Benatar singing