Tight integration of acoustic models with LLMs for ASR is ablated against shallow fusion across label units, fine-tuning strategies, LLM sizes, and joint CTC decoding to mitigate hallucinations.
Adapter.The adapter can optionally further downsample the encoder output, and then optionally performs a linear transfor- mation to match the decoder input dimension
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
LLMs and Speech: Integration vs. Combination
Tight integration of acoustic models with LLMs for ASR is ablated against shallow fusion across label units, fine-tuning strategies, LLM sizes, and joint CTC decoding to mitigate hallucinations.