Warming up adapters on linguistically related source languages via multitask learning or MAML improves low-resource ASR adaptation of frozen SSL models by up to 28% relative CER/PER over adapter-only PEFT.
Dataset We evaluate our solution using ML-SUPERB [15], a benchmark for multilingual ASR with speech SSL models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SD 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
How to Learn a New Language? An Efficient Solution for Self-Supervised Learning Models Unseen Languages Adaption in Low-Resource Scenario
Warming up adapters on linguistically related source languages via multitask learning or MAML improves low-resource ASR adaptation of frozen SSL models by up to 28% relative CER/PER over adapter-only PEFT.