anjalyjayakrishnan/test

Snow Mountain数据集包含11种印度语言的圣经音频录音（以.mp3格式）和相应的文本。录音由母语者在录音室环境中完成，每种语言在数据集中只有一个说话者。这些语言大多集中在印度北部的喜马偕尔邦附近，并且由于与印地语相关，它们都使用天城文进行转录。该数据集已用于自动语音识别任务的实验，但也可用于语音领域的其他应用，如说话人识别、语言识别，甚至作为预训练的无标签语料库。The Snow Mountain Dataset contains Bible audio recordings in .mp3 format and their corresponding transcript texts. The audio recordings were collected from native speakers in a professional studio environment, with exactly one unique speaker per language included in the dataset. Most of these languages are concentrated near Himachal Pradesh in northern India, and all utilize the Devanagari script for transcription due to their linguistic proximity to Hindi. This dataset has been employed in experiments for automatic speech recognition (ASR) tasks, but it can also be applied to other speech-domain research applications, such as speaker recognition, language identification, and even as an unlabeled corpus for model pretraining.

原始信息汇总

数据集概述

数据集名称: Snow Mountain

语言:

• Hindi
• Haryanvi
• Bilaspuri
• Dogri
• Bhadrawahi
• Gaddi
• Kangri
• Kulvi
• Mandeali
• Kulvi Outer Seraji
• Pahari Mahasui

支持的任务:

• 自动语音识别
• 说话人识别
• 语言识别

数据集大小:

• 总时长: 71.41小时 (hi配置) + 27.41小时 (bgc配置)

数据集结构:

• 配置 hi:
  • 特征:
    • Unnamed: int64
    • sentence: string
    • path: string
  • 分割:
    • train_500: 400个样本
    • val_500: 100个样本
    • train_1000: 800个样本
    • val_1000: 200个样本
    • test_common: 500个样本
• 配置 bgc:
  • 特征:
    • Unnamed: int64
    • sentence: string
    • path: string
  • 分割:
    • train_500: 400个样本
    • val_500: 100个样本
    • train_1000: 800个样本
    • val_1000: 200个样本
    • test_common: 500个样本

数据集创建:

• 源数据: 《圣经》录音由母语者在一个工作室环境中完成。
• 许可证: Creative Commons Attribution-ShareAlike 4.0 International Public License (CC BY-SA 4.0)

引用信息:

@inproceedings{Raju2022SnowMD, title={Snow Mountain: Dataset of Audio Recordings of The Bible in Low Resource Languages}, author={Kavitha Raju and V. Anjaly and R. Allen Lish and Joel Mathew}, year={2022} }