Failed in reproduce result in PromptASR

[tienanh28122000]

Hi everyone, I've downloaded the pre-trained models of PromptASR to reproduce the result that has been report. However, when I reproduce with the recipes, the result between using prompt (use_pre_text=True) and not using promt (use_pre_text=False) is the same. For example, both approach outputs the transcription "Bessy took a long and feverish draught, and then fell back and shut her eyes.". Therefore, is there any bug in the code or I was reproduce wrong?

[marcoyang1998]

Hi, which model were you using? And could you please show me your inference code?

[tienanh28122000]

Hi @marcoyang1998, thank you for your reply. I've downloaded model from your hugging face and used your inference code in this link: #1250 (comment). However, for both model (utterence-level model and word-level model), I have the same result between using prompt and not using prompt. It's not like your result report in that link.