search

usefulsensors / moonshine

Fast and accurate automatic speech recognition (ASR) for edge devices
MIT License
2.25k stars 98 forks source link

Pretraining method #42

Open oggyfaker opened 3 weeks ago

oggyfaker commented 3 weeks ago

Hi i am really curious about the loss function you use for train pretrain-weight (CTC or Cross-Entropy loss ?) . It not be mentioned in paper. Hope to see your answer !