{"ID":5937873,"CreatedAt":"2026-07-07T03:14:33.014478982Z","UpdatedAt":"2026-07-09T02:45:08.403619813Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2607.04619","arxiv_id":"2607.04619","title":"CARD: Cross-component Audio Representation Distillation for Encoder-Free Audio Captioning","abstract":"Modern automated audio captioning systems pair a frozen audio encoder with a large language model (LLM) via a trainable projector, incurring the encoder's inference cost and bottlenecking the model through its fixed acoustic features. We present CARD, an encoder-free audio captioning model that removes the encoder at inference: a 13.2M projector feeds a frozen LLM with merged LoRA adapters, while the teacher used to train it is discarded. CARD distills a pretrained audio teacher (CLAP-HTSAT) into the model, but rather than injecting it into the LLM alone, it routes the teacher's representations across components: perceptual stages to the projector and semantic stages to the LLM. This placement improves CIDEr-D by +12.18 over an LLM-only distilled model on AudioCaps and by +5.21 on Clotho, reaching 55.4 against a 66.4 encoder-kept upper bound with no encoder at inference, showing that where a teacher's knowledge is placed matters as much as its presence.","short_abstract":"Modern automated audio captioning systems pair a frozen audio encoder with a large language model (LLM) via a trainable projector, incurring the encoder's inference cost and bottlenecking the model through its fixed acoustic features. We present CARD, an encoder-free audio captioning model that removes the encoder at i...","url_abs":"https://arxiv.org/abs/2607.04619","url_pdf":"https://arxiv.org/pdf/2607.04619v1","authors":"[\"Ganesh Pavan Kartikeya Bharadwaj Kolluri\",\"Yuchen Zhang\",\"Michael Kampouridis\",\"Ravi Shekhar\"]","published":"2026-07-06T02:58:47Z","proceeding":"cs.SD","tasks":"[\"cs.SD\",\"cs.CL\"]","methods":"[\"Large Language Model\",\"Language Model\",\"LoRA\"]","has_code":false}