More Evaluation Metrics: Articulatory Reconstruction Loss (ARL) #2
Description
We currently only have Phoneme Error Rate (PER) and Weighted Phonetic Error Rate/Feature Error Rate (WPER/FER) for evaluating Speech2IPA models (see metrics.py). Zongli Ye et al. introduce ARL loss to evaluate their Speech2IPA model using the AAI model by Cho et al.:
Would be great to also have this eval metric.