audio-encoder representations

Audio LLMs can now detect unreliable transcriptions

Researchers have developed a method for Audio Large Language Models (LLMs) to recognize when they cannot reliably transcribe user input. Current methods, including speech quality predictors and generation uncertainty, o…