speedykom-group/pokoot-east-speech-dataset

Pokoot East语音数据集是一个用于支持肯尼亚南方尼罗特语（Kalenjin）中Pokoot East方言的语音数据集。数据集包含音频和转录文本，音频为WAV格式，16 kHz采样率，单声道；转录文本为UTF-8编码。数据集分为训练集（600个样本，60%）、验证集（200个样本，20%）和测试集（200个样本，20%）。数据来源于Global Recordings Network的GRN圣经叙述，通过静音检测分段。转录由facebook/mms-1b-all模型的Kpz适配器自动生成，建议人工审查和校正。数据集由Speedykom创建，旨在推动非洲弱势语言的语音技术发展。

The Pokoot East Speech Dataset is a speech dataset for the Pokoot East dialect, a Southern Nilotic (Kalenjin) language spoken in Kenya. It includes audio in WAV format (16 kHz, mono) and UTF-8 transcripts. The dataset is split into train (600 samples, 60%), validation (200 samples, 20%), and test (200 samples, 20%) sets. The audio is sourced from GRN Bible narratives by Global Recordings Network, segmented via silence detection. Transcriptions were auto-generated using the Kpz adapter of the `facebook/mms-1b-all` model, with manual review recommended. Created by Speedykom, this dataset aims to advance speech technology for underserved African languages.