#audiocaptioning
Mamba-2, a state‑space language model, achieves audio captioning with fewer parameters than larger transformer LLMs, using LoRA fine‑tuning and connectors. Read more: https://getnews.me/mamba-2-advances-audio-captioning-via-design-space-exploration/ #mamba2 #audiocaptioning #statespacemodel
September 22, 2025 at 10:46 AM
Researchers introduced an audio captioning framework that combines RLHF with a CLAP‑based reward model, and human evaluations show its captions are preferred over baseline systems. Read more: https://getnews.me/audio-captioning-aligned-with-human-preferences-via-rlhf/ #audiocaptioning #rlhf #clap
September 20, 2025 at 5:37 AM
Zero‑shot audio captioning using audio‑CLIP and an LLM achieved a 35% gain, with the NLG mean score on WavCaps rising from 4.7 to 7.3 via the MAGIC keyword prompt. Read more: https://getnews.me/zero-shot-audio-captioning-boosted-by-magic-keyword-prompts-and-clip/ #audiocaptioning #magic
September 18, 2025 at 8:37 AM