Beyond WER: Entity and Disfluency Recall in Accented Conversational ASR
Beyond WER: Entity and Disfluency Recall in Accented Conversational ASR
超越词错误率(WER):口音对话式语音识别中的实体与非流利词召回
ASR systems optimised for Word Error Rate (WER) often miss named entities and filled pauses in accented conversational English, both critical for language-learning feedback. 针对词错误率(WER)进行优化的自动语音识别(ASR)系统,往往会遗漏带口音的对话式英语中的命名实体和填充停顿,而这两者对于语言学习反馈至关重要。
We present a three-stage pipeline for speakers from India, Indonesia, and Latin America: (1) heuristic SQL filters curating entity-rich training data at 2.8x the entity density of random sampling, (2) regional LoRA adapters fine-tuned on Qwen2.5-Omni-3B producing both verbatim and corrected transcripts in a single forward pass, and (3) a six-category error taxonomy validated by an LLM-based judge (83.8% agreement, 210 human-labelled samples). 我们针对来自印度、印度尼西亚和拉丁美洲的说话人提出了一套三阶段流水线:(1) 使用启发式 SQL 过滤器筛选出实体丰富的训练数据,其实体密度是随机采样的 2.8 倍;(2) 基于 Qwen2.5-Omni-3B 微调区域性 LoRA 适配器,在单次前向传播中同时生成逐字转录和修正后的转录;(3) 建立了一个包含六类错误的分类体系,并由基于大语言模型的裁判进行验证(一致性达 83.8%,基于 210 个经人工标注的样本)。
The pipeline achieves 80-85% entity recall (up from 53-55%), 76-86% filler recall (up from <5%), and 6-10% WER across 6k test utterances, outperforming Whisper and a commercial ASR on entity recall while matching a zero-shot 30B model with 10x fewer parameters. 该流水线在 6000 条测试语句中实现了 80-85% 的实体召回率(从 53-55% 提升)、76-86% 的填充词召回率(从 <5% 提升)以及 6-10% 的 WER。在实体召回方面,该模型优于 Whisper 和某商业 ASR 系统,同时在参数量减少 10 倍的情况下,表现与零样本(zero-shot)的 30B 模型相当。
Paired bootstrap tests confirm that curation alone accounts for 2.8-4.2 pp of entity recall gain (p<0.0001). 配对自助法(Paired bootstrap)测试证实,仅数据筛选这一项就贡献了 2.8-4.2 个百分点的实体召回率提升(p<0.0001)。