We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:


References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: Sequence Model with Self-Adaptive Sliding Window for Efficient Spoken Document Segmentation

Abstract: Transcripts generated by automatic speech recognition (ASR) systems for spoken documents lack structural annotations such as paragraphs, significantly reducing their readability. Automatically predicting paragraph segmentation for spoken documents may both improve readability and downstream NLP performance such as summarization and machine reading comprehension. We propose a sequence model with self-adaptive sliding window for accurate and efficient paragraph segmentation. We also propose an approach to exploit phonetic information, which significantly improves robustness of spoken document segmentation to ASR errors. Evaluations are conducted on the English Wiki-727K document segmentation benchmark, a Chinese Wikipedia-based document segmentation dataset we created, and an in-house Chinese spoken document dataset. Our proposed model outperforms the state-of-the-art (SOTA) model based on the same BERT-Base, increasing segmentation F1 on the English benchmark by 4.2 points and on Chinese datasets by 4.3-10.1 points, while reducing inference time to less than 1/6 of inference time of the current SOTA.
Comments: accepted by ASRU 2021
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2107.09278 [cs.CL]
  (or arXiv:2107.09278v2 [cs.CL] for this version)

Submission history

From: Qinglin Zhang [view email]
[v1] Tue, 20 Jul 2021 06:44:13 GMT (783kb,D)
[v2] Sat, 9 Oct 2021 06:38:31 GMT (788kb,D)

Link back to: arXiv, form interface, contact.