chocobearz/BERSt

BERSt数据集是一个用于自动语音识别（ASR）和语音情感识别（SER）任务的数据集。它包含了4526个单句录音，时长约为3.75小时，由98名专业演员在19种手机位置、7种情感类别和3种语音强度级别下录制。数据集涵盖了多种区域和非母语英语口音，以及13个无意义短语，以确保覆盖所有英语音素。数据收集于家庭环境中，使用智能手机麦克风，参与者来自全球各地，代表了不同的区域口音和非母语英语口音。数据集提供了训练、测试和验证集，且没有跨集的说话者重叠。元数据包括演员数量、性别分布、日常语言和母语的统计信息。

The BERSt dataset is a collection of 4,526 single phrase recordings (~3.75 hours) designed for Automatic Speech Recognition (ASR) and Speech Emotion Recognition (SER) tasks. It features recordings from 98 professional actors in 19 phone positions, with 7 emotion classes and 3 vocal intensity levels. The dataset includes varied regional and non-native English accents, as well as 13 nonsense phrases covering all English phonemes. The data was collected in home environments using smartphone microphones, with participants from around the globe representing diverse regional accents and non-native English speakers. The dataset provides train, test, and validation splits with no speaker crossover between splits. Metadata includes details on actor count, gender distribution, daily language, and first language statistics.