sulabhkatiyar/ne-asr-mjw-aug
收藏资源简介:
这是一个针对Karbi语言(ISO 639-3代码:mjw)的增强自动语音识别(ASR)数据集。Karbi是一种在印度阿萨姆邦使用的藏缅语系有声调语言。数据集基于原始Vaani项目录音,通过速度扰动增强(0.9倍和1.1倍速度)将训练样本从508个扩展到1,524个(3倍增强),但未应用音高偏移以保留声调语言的词汇意义。数据集包含训练、验证和测试分割,音频为16kHz单声道WAV格式,存储为Parquet文件,并包含音频、文本、语言和增强标签等特征。许可证为CC-BY-NC 4.0。
Augmented automatic speech recognition dataset for Karbi (mjw), a Tibeto-Burman tonal language spoken in Assam, India. The dataset is enhanced from original Vaani project recordings via speed perturbation (0.9x and 1.1x speed) to expand training samples from 508 to 1,524 (3x augmentation), without pitch shift to preserve lexical tone contrasts. It includes train, validation, and test splits, with audio in 16kHz mono WAV format stored as Parquet files, featuring audio, text, language, and augmentation labels. Licensed under CC-BY-NC 4.0.




