duplexio/emilia-yodas-en-speaker-embeddings

该数据集包含了Emilia-YODAS英语子集的预计算说话人嵌入向量。每一行将一个Emilia-YODAS样本ID映射到从对应音频中提取的一个说话人嵌入向量。数据集来源为`amphion/Emilia-Dataset`的Emilia-YODAS英语子集，使用`Qwen/Qwen3-TTS-12Hz-1.7B-Base`模型生成嵌入向量，嵌入形状为`(2048,)`，数据类型为`float16`，共4,516,833行，仅包含训练分割。数据以Arrow格式存储，包含`id`（源样本ID）和`speaker_embedding`（2048维说话人嵌入向量）两列。这些嵌入向量旨在用于语音模型预训练、说话人条件化、说话人检索或需要Emilia-YODAS英语音频的每话语说话人表示的分析工作流程。注意，数据集仅包含嵌入向量，不包含原始音频或转录文本，嵌入质量取决于上游音频质量和嵌入模型，且不应视为已验证的说话人身份。

This dataset contains precomputed speaker embeddings for the English subset of Emilia-YODAS. Each row maps an Emilia-YODAS sample ID to one speaker embedding extracted from the corresponding audio. Source dataset: `amphion/Emilia-Dataset`, source subset: Emilia-YODAS English, embedding model: `Qwen/Qwen3-TTS-12Hz-1.7B-Base`, embedding shape: `(2048,)`, embedding dtype: `float16`, rows: 4,516,833, split: `train`, additional local-access format: `arrow/`, columns: `id` (source Emilia-YODAS sample ID) and `speaker_embedding` (2048-dimensional speaker embedding). These embeddings are intended for speech model pretraining, speaker conditioning, speaker retrieval, or analysis workflows that need per-utterance speaker representations for Emilia-YODAS English audio. Limitations: The dataset contains embeddings only, not the original audio or transcripts; embedding quality depends on the upstream audio quality and on the embedding model; these embeddings should not be treated as verified speaker identities.