Korean speech error correction improves call center transcript accuracy
Leveraging Fine-grained Error Correction in Korean Speech Recognition for Consultation Services
Computation and Language
Summary
Accurate transcripts from speech recognition are hard to get, especially for languages like Korean with few resources. The authors created a large dataset from real call center conversations to help fix errors in these transcripts using just text. They designed a new method that first finds mistakes word by word, then carefully corrects the errors by looking at the surrounding conversation. Their approach works better than general language models, helping make automatic Korean call transcripts more reliable without needing the original audio.
What this means in practice
- •For call center operations teams: Improve the accuracy of Korean automatic transcripts for customer calls without accessing call audio.
- •For language technology developers: Develop better Korean speech recognition correction tools using a large dialogue-level dataset and context-aware models.
Authors
Yonghyun Jun, Jimin Lee, Hwan Chang, Dongho Shin, Seolah Kim, Hwanhee Lee
Abstract
Automatic Speech Recognition (ASR) technology is fundamental to customer service automation and large-scale transcription. However, even advanced ASR models exhibit inevitable errors in complex real-world environments such as call center conversations. When privacy restrictions preclude audio access, error correction must rely on text-based post-editing. Existing text-only approaches face significant challenges in low-resource languages, mainly due to a critical scarcity of annotated corpora and tailored correction methodologies. For Korean, this resource gap is particularly pronounced, as existing resources are predominantly designed for ASR training rather than text-based error correction. To address this, we introduce DasanCallDial, the first large-scale Korean benchmark dataset specifically curated for dialogue-level ASR error correction. Derived from genuine call center interactions, it comprises 1,974 dialogues with 115,460 utterances. Leveraging this resource, we propose Detector-Gated Contextual Span Correction (DCSC), a text-only post-editing framework for error-sparse Korean speech recognition transcripts. DCSC combines an encoder-based detector that first performs token-level error detection, followed by a language model-based corrector trained to rectify fine-grained span-level errors. Additionally, we employ dialogue-level context augmentation to enable the model to leverage discourse history for disambiguation. By employing multi-level granularity, our method achieves state-of-the-art performance, effectively overcoming the limitations of general LLMs in low-resource settings.