I started looking a bit into Confidence Scores / Self-Training for Speech Recognition for models like Wav2Vec2. The most reasonable way of doing so is to do it on a per-word level basis. With the new output_word_offsets=True it’s quite easy to retrieve the logits scores of the predicted words. E.g…

Confidence Scores / Self-Training for Wav2Vec2 / CTC models

patrickvonplaten April 21, 2022, 11:14am 2

Topic		Replies	Views
Wav2Vec2: loss growing in training and validation after few epochs Models	5	1453	February 1, 2024
Original and re-loaded model are not the same Beginners	0	315	August 14, 2021
Wav2Vec2: fix growing training and validation loss after few epochs Models	5	1988	January 27, 2022
Fine-tune wav2vec2-large-xlsr-53 for one epoch 🤗Transformers	0	327	January 11, 2022
Different evaluation results during and after training: Wav2Vec2 finetuning 🤗Transformers	0	294	January 25, 2023