Firstly, thank you for the excellent project. I am also working on integrating the Hugging Face Wav2Vec2 model with WFST for shallow fusion and context biasing. I used K2 to create an HLG graph, but when I decoded it with the Wav2Vec model, the output was weird. Following your advice, I added a blank_score (for me it was 15), and everything went perfectly.
Could you please explain why you added blank_score to your code and why it is set to 0 by default?
Lastly, I want to thank you again because your code helped me solve my problem. I hope you have a good day.
Firstly, thank you for the excellent project. I am also working on integrating the Hugging Face Wav2Vec2 model with WFST for shallow fusion and context biasing. I used K2 to create an HLG graph, but when I decoded it with the Wav2Vec model, the output was weird. Following your advice, I added a blank_score (for me it was 15), and everything went perfectly.
Could you please explain why you added blank_score to your code and why it is set to 0 by default?
Lastly, I want to thank you again because your code helped me solve my problem. I hope you have a good day.