Two-step phoneme-based ASR with LLM-P2G decoding, using noisy-phoneme augmentation and randomized top-K marginalized training, reduces WER on Polish and German versus WFST decoding.
Dataset Experiments are conducted on the CommonV oice (CV) dataset [22], version 11.0 (released September 2022)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SD 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
LLM-based phoneme-to-grapheme for phoneme-based speech recognition
Two-step phoneme-based ASR with LLM-P2G decoding, using noisy-phoneme augmentation and randomized top-K marginalized training, reduces WER on Polish and German versus WFST decoding.