PTC-Bias: Phoneme-Level Temporal Competition for Bias Retrieval and Post-Decoding Correction in Speech LLMs
PTC-Bias is proposed, a two-stage framework based on phoneme-level temporal competition that reduces near-homophone and word-segmentation errors while preserving correct transcriptions and shows consistent gains across two SpeechLLMs and bias lists of up to 2000 words.
Zhi-Qi Ai, Han Cheng, Shi-Yi Mu et al.
· 0 citations