{"version":1,"type":"story","url":"https://digestai.news/story/qwen2-5-omni-3b-adapters-boost-entity-recall-in-accented-conversationa","json":"https://digestai.news/story/qwen2-5-omni-3b-adapters-boost-entity-recall-in-accented-conversationa.json","markdown":"https://digestai.news/story/qwen2-5-omni-3b-adapters-boost-entity-recall-in-accented-conversationa.md","slug":"qwen2-5-omni-3b-adapters-boost-entity-recall-in-accented-conversationa","headline":"Qwen2.5-Omni-3B adapters boost entity recall in accented conversational ASR","summary":"A new three‑stage pipeline targets accented English speech from India, Indonesia and Latin America, where standard ASR systems optimized for word error rate often miss named entities and filler words. First, heuristic SQL filters curate training data that is 2.8 × richer in entities than random samples. Second, regional LoRA adapters fine‑tuned on the Qwen2.5‑Omni‑3B model generate both verbatim and corrected transcripts in a single forward pass. Third, a six‑category error taxonomy is evaluated by an LLM‑based judge, achieving 83.8 % agreement on 210 human‑labelled samples.\n\nOn a 6 k‑utterance test set the pipeline reaches 80‑85 % entity recall (up from 53‑55 %) and 76‑86 % filler recall (up from under 5 %), while keeping word error rate between 6 % and 10 %. It outperforms Whisper and a commercial ASR on entity recall and matches a zero‑shot 30 B‑parameter model with ten‑fold fewer parameters. Paired bootstrap tests attribute 2.8‑4.2 percentage‑point gains in entity recall to data curation alone (p < 0.0001).","keyPoints":["Heuristic SQL filters produce training data with 2.8× higher entity density than random sampling.","LoRA adapters fine‑tuned on Qwen2.5‑Omni‑3B raise entity recall to 80‑85% and filler recall to 76‑86% on accented speech.","Pipeline matches a zero‑shot 30B model’s performance with only 3B parameters and 10× fewer resources."],"whyItMatters":"Improving entity and filler detection in accented ASR helps language‑learning tools provide more accurate feedback, expanding accessibility for non‑native speakers.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":[],"models":["Qwen2.5-Omni-3B","Whisper"],"people":[]},"firstPublishedAt":"2026-09-21T04:00:00Z","updatedAt":"2026-09-21T04:00:00Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"arXiv cs.CL","title":"Beyond WER: Entity and Disfluency Recall in Accented Conversational ASR","url":"https://arxiv.org/abs/2609.20828","publishedAt":"2026-09-21T04:00:00Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[],"thread":{"title":"AI Whisper Advances Multilingual Speech Recognition","url":"https://digestai.news/thread/token-merging-boosts-whisper-efficiency-across-16-languages-with-minimal","storyCount":2},"cite":{"text":"Digest AI, \"Qwen2.5-Omni-3B adapters boost entity recall in accented conversational ASR\", 21 September 2026, https://digestai.news/story/qwen2-5-omni-3b-adapters-boost-entity-recall-in-accented-conversationa","publisher":"Digest AI","title":"Qwen2.5-Omni-3B adapters boost entity recall in accented conversational ASR","datePublished":"2026-09-21T04:00:00Z","url":"https://digestai.news/story/qwen2-5-omni-3b-adapters-boost-entity-recall-in-accented-conversationa"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}