Question about the dataset in HW4 about leedl-tutorial HOT 1 CLOSED

The dataset is made from a lot of .pt files, and it is melspectrogram. I have these questions for those files

1. is the dimension of them is mel_len * 40? Is mel_len means the time span, and 40 is the magnitude of dB with 40 different frequency, or the magnitude of frequency with 40 different dB?
2. why the encoder layer expect features in the shape of (length, batch_size, d_model)?

Many thanks for your questions.

1. Yes, the dimension of them is mel_len * 40. 40 is the dimensions of mel mapping. "mel_len" means time span, while "40" means 40 different dB values at different frequencies, thus indicating the amplitude values (dB) at different frequencies.
2. The shape is (batch size, length, mel_len), and maybe d_model equals to mel_len.

from leedl-tutorial.