PTC-Bias: Phoneme-Level Temporal Competition for Bias Retrieval and Post-Decoding Correction in Speech LLMs
A novel framework enhancing speech model accuracy by resolving phoneme-level conflicts in large-scale bias lists.
Contextual biasing improves rare-word recognition in speech large language models (SpeechLLMs), but efficiently exploiting large bias lists remains challenging. We propose PTC-Bias, a two-stage framework based on phoneme-level temporal competition. At the prefill stage, PTC Retri...
ArXiv CS.CL
Source
Generated insight by OnAirToday AI
PTC-Bias represents a significant leap in speech recognition by enabling models to handle massive bias lists without compromising computational efficiency. By resolving temporal conflicts at the phoneme level, this framework enhances accuracy for niche terminology that typically degrades performance in large language models.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.