AI & Audio Processing

2 summarised stories about AI & Audio Processing, each linking back to the original source. Browse all topics →

Thursday, 16 July 2026

NAVER LABS System Re-implementation for the IWSLT 2026 Instruction-Following Task

arXiv cs.CL 18 hours ago

NAVER LABS implemented a speech-to-text instruction-following system for the IWSLT 2026 shared task using SeamlessM4T-v2-large for speech encoding and Qwen3-4B-Instruct as the language model. The system uses a three-stage pipeline with projector alignment, text-only LoRA pre-training, and multimodal merging, supported by 100,000 synthetic instruction-following examples across ten task types. The model achieved COMET 0.781 on English-to-Chinese speech translation and BERTScore-F1 0.346 on English spoken question answering.

AI & Audio Processing AI Code Generation AI Datasets & Annotation AI & Domain Specialization

The Sound of Absence: Audio-Language Embedding Models Struggle with Negation

arXiv cs.CL 18 hours ago

Audio-language embedding models like CLAP fail to properly distinguish between affirmative and negated sound concepts, mapping them to nearly identical representations. A new evaluation framework called NegEval-Audio tested this on AudioCaps and Clotho datasets, finding that models performed far below chance on multiple-choice negation tasks, with the problem persisting even in recent multimodal LLM-based models. Addressing this limitation requires explicit negation-aware training objectives rather than just steering methods.

Adversarial Robustness AI Accuracy & Factuality AI & Audio Processing AI Evaluation