slprl/TinyStress-15K

TinyStress-15K是一个合成语音数据集，包含重音标注，旨在支持研究模型理解句重音（即对特定单词的强调，影响句子的意义）。数据集包括15,000个训练样本和1,000个测试样本，采样率为48kHz。文本来源于TinyStories，并提供了SSML格式化版本用于调整韵律特征。每个样本都有唯一的标识符、原始样本索引、句子索引、音频波形、文本转录、SSML格式文本、重音索引、元数据（包括性别、语言代码和声音名称）和单词开始时间戳。

TinyStress-15K is a synthetic speech dataset with stress annotations designed to support research on models that understand sentence stress, which is the emphasis on specific words that affect the meaning of a sentence. The dataset includes 15,000 training samples and 1,000 testing samples, with a sampling rate of 48kHz. The texts are derived from TinyStories and an SSML-formatted version is provided to manipulate prosodic features. Each sample contains a unique identifier, the index of the original TinyStories sample, sentence index, audio waveform, text transcription, SSML-formatted text, indices of words with emphasis, metadata including gender, language code, and voice name, as well as word start timestamps.