Mandarin-English Conversational Telephone Speech Corpus

本论文介绍了一个针对普通话-英语对话电话语音的语料库，名为'Mandarin-English Conversational Telephone Speech Corpus'，由雷神BBN创建。该数据集包含123.5小时的普通话对话电话语音数据，来源于CallHome Mandarin Chinese和HKUST Mandarin Telephone Speech数据集。数据集的创建过程涉及由双语注释者提供的英语翻译，确保了语音数据的准确性和适用性。该数据集主要用于构建和改进语音翻译系统，特别是在提高机器翻译模型在特定领域的性能方面具有重要价值。

This paper introduces a corpus for Mandarin-English conversational telephone speech, named 'Mandarin-English Conversational Telephone Speech Corpus', developed by Raytheon BBN. Comprising 123.5 hours of Mandarin conversational telephone speech data, the corpus is sourced from the CallHome Mandarin Chinese and HKUST Mandarin Telephone Speech datasets. During the construction of this corpus, bilingual annotators provided English translations, which ensures the accuracy and applicability of the speech data. This corpus is primarily used for building and improving speech translation systems, and is particularly valuable for enhancing the performance of machine translation models in specific domains.

搜集汇总

背景与挑战

背景概述

该数据集是一个普通话-英语对话电话语音语料库，包含123.5小时的语音数据，来源于CallHome Mandarin Chinese和HKUST Mandarin Telephone Speech数据集，并由双语注释者提供英语翻译。它主要用于构建和改进语音翻译系统，以提升机器翻译模型在特定领域的性能。