[Feature Request]: AdaMER-CTC for ASR task training

[tomschelsen]

🚀 The feature

Adaptive Maximum Entropy Regularization, as described in : https://arxiv.org/abs/2403.11578

Solution outline

Switchable training parameter (classic CTC versus AdaMER-CTC) with two additional hyperparameters β (as in the paper) and ... α ? (not named in the paper but which would be the coef between target entropy and target length : α = H / U

Additional context

No response