iiinova — where your digital self keeps evolving

Beomseok Lee @beomseok-lee.bsky.social · 29/10/2025

Can we make Speech LLMs actually think as they listen? 👂💭 This fascinating work applies CoT inspired by human “thinking while listening”, training models to find the inflection point when reasoning starts. 📄 arxiv.org/abs/2510.07497

arxiv.org

Can Speech LLMs Think while Listening?

Recent advances in speech large language models (speech LLMs) have enabled seamless spoken interactions, but these systems still struggle with complex reasoning tasks. Previously, chain-of-thought (Co...

000

Beomseok Lee @beomseok-lee.bsky.social · 11/06/2025

Speech-language models show promise in multimodal tasks—but how well are speech & text actually aligned? 🤔 This paper arxiv.org/abs/2505.19937 proposes a new metric to measure layer-wise correlation between the two, with a focus on SLU tasks. 🔍🗣️📄

arxiv.org

ALAS: Measuring Latent Speech-Text Alignment For Spoken Language Understanding In Multimodal LLMs

Large Language Models (LLMs) are widely used in Spoken Language Understanding (SLU). Recent SLU models process audio directly by adapting speech input into LLMs for better multimodal learning. A key c...

000

Beomseok Lee @beomseok-lee.bsky.social · 03/04/2025

Should speech come before the instruction text, or should the instruction text come first in a speech-language model? Find out the best positioning for speech and text—and the novel adapter that aligns speech and text modalities! arxiv.org/abs/2412.01145

arxiv.org

AlignFormer: Modality Matching Can Achieve Better Zero-shot Instruction-Following Speech-LLM

Integrating speech into LLM (speech-LLM) has gaining increased attention recently. The mainstream solution is to connect a well-trained speech encoder and LLM with a neural adapter. However, the lengt...

020