4600小时全双工普通话自然对话语音数据

全双工普通话自然对话语音数据，基于常见主题对话录制。此数据集标注了文本内容、句时间戳、说话人身份、性别等多种属性，准确性高，为语音识别相关研究及应用提供了丰富的资源，经多家AI公司验证：有助于模型面对真实世界的多样性时能够表现出色。我们严格遵循数据保护法规和隐私规定，确保数据采集、存储和使用的过程中维护用户的隐私和合法权益，所有数据均遵循GDPR, CCPA, PIPL。Full-duplex Mandarin natural conversational speech dataset recorded based on common daily conversation topics. This dataset is annotated with multiple attributes including transcript content, sentence-level timestamps, speaker identity, gender and others, with high annotation accuracy. It provides abundant resources for research and applications related to speech recognition. Verified by multiple AI companies, this dataset helps models deliver excellent performance when facing the diversity of real-world scenarios. We strictly comply with data protection regulations and privacy rules, ensuring that user privacy and legitimate rights and interests are protected throughout the entire process of data collection, storage and usage. All data in this dataset adheres to GDPR, CCPA and PIPL.

搜集汇总

背景与挑战

背景概述

该数据集是一个4600小时全双工普通话自然对话语音数据，基于常见主题录制，标注了文本内容、时间戳、说话人身份和性别等多种属性，准确率高（字准确率99%）。它覆盖约3600名录音人，男女比例均衡，适用于语音识别和声纹识别等应用，具有大规模、高质量和真实世界多样性的特点。