ClovaCall

我们在11,000多人的目标导向对话场景下引入了一种新的基于韩国呼叫的大规模语音语料库，即Clova呼叫语料库 (ClovaCall)。ClovaCall的原始数据集包括餐馆预订域中的大约112,000对短句及其对应的口语。我们通过对两个最先进的ASR模型进行深入实验来验证数据集的有效性。

We introduce a novel large-scale Korean call-based speech corpus, named ClovaCall (Clova Call Corpus), for goal-oriented conversational scenarios involving over 11,000 participants. The original dataset of ClovaCall contains approximately 112,000 pairs of short sentences and their corresponding spoken utterances within the restaurant reservation domain. We conduct in-depth experiments on two state-of-the-art ASR models to validate the effectiveness of this corpus.

搜集汇总

背景与挑战

背景概述

ClovaCall是一个大规模的韩语语音语料库，专注于目标导向的对话场景，特别是餐馆预订领域，包含约112,000对短句及其对应的口语，适用于自动语音识别研究。数据集由多个机构联合发布，包括香港科技大学、NAVER和Hankuk University of Foreign Studies，发布时间为2020年。