Brief

A new research paper analyzes how errors in Korean speech recognition impact the performance of large language models (LLMs) in spoken question answering (SQA). The study found that the degradation caused by speech recognition errors is consistent across different LLMs, suggesting that the information loss at the speech recognition stage is the primary driver of performance decline. The research also identified single-character errors in Korean transcriptions as a unique vulnerability that can alter the intended question and degrade QA accuracy. An auxiliary comparison indicated that large audio language models may offer a more robust solution by directly processing audio input, potentially mitigating issues caused by transcription errors. AI

Researchers have developed VocalParse, a new model for transcribing singing voices that utilizes a Large Audio Language Model (LALM). This model addresses limitations in current systems by jointly modeling lyrics, melody, and text-note alignments through an interleaved prompting formulation. VocalParse also employs a Chain-of-Thought strategy to first decode lyrics, which helps maintain structural integrity and improve transcription accuracy, achieving state-of-the-art results on various singing datasets. AI