Arabic Call Benchmark Dataset

Arabic Call Benchmark Dataset是由Maqsam公司创建的一个专注于阿拉伯语电话通话自动语音识别（ASR）的基准数据集。该数据集包含来自13个不同国家的电话通话记录，涵盖教育、娱乐和电子商务等多个领域，特别关注客户支持和销售职位。数据集标准化为16kHz采样率，单声道，16位精度，包含从清晰到嘈杂环境的不同音频质量。通过41名注释者和13名审阅者的严格手工转录，形成了132小时的高质量标注语音数据。此数据集旨在为ASR系统提供一个全面的测试平台，以解决阿拉伯语电话通话中的语音识别挑战，推动ASR技术在阿拉伯语区域的发展和应用。

Arabic Call Benchmark Dataset is a benchmark dataset focused on automatic speech recognition (ASR) for Arabic telephone calls, created by Maqsam. This dataset contains telephone call records from 13 different countries, covering multiple fields such as education, entertainment and e-commerce, with particular focus on customer support and sales roles. The dataset is standardized to 16kHz sampling rate, mono channel, 16-bit precision, and includes varying audio qualities ranging from clear to noisy environments. It consists of 132 hours of high-quality annotated speech data obtained through strict manual transcriptions conducted by 41 annotators and 13 reviewers. This dataset aims to provide a comprehensive test platform for ASR systems to address the speech recognition challenges in Arabic telephone calls, and promote the development and application of ASR technology in the Arabic-speaking region.