Application on ASR

[chlorane]

I am interested in your Audio Spectrogram Transformer, and I want to try to apply it om our project.
Our project focuses on ASR and wants to take the audio spectrogram(in our data, 43D) as input and produce its text as output. We tried several other transformers(from hugging face) but failed, because their input is audio wave form.
I'm wondering whether your AST model can take the spectrogram directly as input. We checked the hugging face version and found they are using wave form directly. If your AST model can do this, do we need to use certain parameter/flags, or change the neural network structures? I think add a decoder is necessary