Automatic Speech Recognition on Ikema romaji (test)
14.8CERxls-r-300m
Evaluation results
| Method | Links | ||
|---|---|---|---|
| xls-r-300mArchitecture=wav2vec2-xls-r-300m, Number of parameters=300M, Pre-training languages=128, Batch size=16, Epochs=50, Learning rate=0.00032026.03 | 14.8 | 64.99 | |
| mms-1bArchitecture=mms-1b, Number of parameters=1B, Pre-training languages=1,406, Batch size=4, Epochs=50, Learning rate=0.00032026.03 | 18.98 | 73.87 | |
| xls-r-1bArchitecture=wav2vec2-xls-r-1b, Number of parameters=1B, Pre-training languages=128, Batch size=4, Epochs=50, Learning rate=0.00032026.03 | 20.33 | 76.73 |