数据堂534小时台湾口音普通话自然对话手机采集语音数据

534小时台湾口音普通话自然对话手机采集语音数据，基于30余个常见主题来模拟录制。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，由来自台湾地区的人员录制，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。

This is a 534-hour dataset of natural dialogues in Taiwan-accented Mandarin, collected via mobile recording devices. The recordings were simulated based on over 30 common conversational topics. This dataset is annotated with comprehensive attributes including transcriptions, sentence-level timestamps, speaker identities, gender information and other relevant metadata, and was recorded by individuals from Taiwan with high annotation accuracy. It serves as a valuable resource for research and applications related to speech recognition, and has been validated by multiple AI companies to help models achieve excellent performance when facing diverse real-world scenarios. We strictly comply with data protection regulations and privacy norms, ensuring that user privacy and legitimate rights and interests are fully protected throughout the entire process of data collection, storage and utilization. All datasets adhere to GDPR, CCPA and PIPL.

搜集汇总

背景与挑战

背景概述

该数据集包含534小时通过手机采集的台湾口音普通话自然对话语音，基于30多个常见话题，并标注了文本、时间戳和说话人信息。数据格式为16kHz WAV，由480名台湾参与者录制，准确率达95%，适用于语音识别研究，版权归DataTang所有。