Brief

Researchers have introduced Raon-Speech, a 9-billion parameter speech language model capable of understanding, answering, and generating speech in English and Korean. This model, trained on over 1.38 million hours of curated speech and text data, outperforms similarly sized audio foundation models on speech-centric tasks while maintaining strong text-based question-answering abilities. An extension, Raon-SpeechChat, further enhances real-time, full-duplex conversation capabilities through additional training on dialogue data, demonstrating strengths in turn-taking and interruption sensitivity. AI

Researchers have introduced AffectVerse, a new multimodal model designed for affective computing that integrates temporal prediction into its reasoning process. Unlike previous models that treated emotion recognition statically, AffectVerse uses an Emotion World Module (EWM) to imagine and predict future affective states based on past multimodal cues. This predictive capability, achieved through cross-modal temporal imagination and belief aggregation, reportedly improves performance by at least 2.57% across nine benchmarks. AI