OpenWhistle-1.0-Pretraining
收藏资源简介:
该数据集包含高采样率(96kHz)的音频记录及其相关元数据,适用于音频处理和分析任务。数据集提供两种配置:默认配置和审查样本配置。每条记录包含音频数据、开始时间、结束时间、持续时间、记录年份和水听器标识符等字段。默认配置包含28,410个训练样本和3,370个验证样本,总大小约78.99GB;审查样本配置包含430个训练样本和50个验证样本,总大小约1.28GB。该数据集适用于声学研究、环境声音分析或机器学习模型的训练与验证。
This dataset contains high sample rate (96kHz) audio recordings and their associated metadata, suitable for audio processing and analysis tasks. The dataset provides two configurations: default configuration and reviewed samples configuration. Each record includes fields such as audio data, start time, end time, duration, recording year, and hydrophone identifier. The default configuration contains 28,410 training samples and 3,370 validation samples, with a total size of approximately 78.99GB; the reviewed samples configuration contains 430 training samples and 50 validation samples, with a total size of approximately 1.28GB. This dataset is suitable for acoustic research, environmental sound analysis, or training and validation of machine learning models.
数据集概述
OpenWhistle 1.0 Pretraining Dataset 是一个用于海豚哨声预训练的无标签音频数据集,由 dolphinteam/OpenWhistle-1.0-Pretraining 提供。
数据集配置
- default:完整的预训练数据集。
- review-sample:一个小型确定性样本子集,便于快速人工审查。
数据特征
- 采样率:96 kHz,单声道音频。
- 字段:
audio(音频)、start_time(开始时间,float32)、end_time(结束时间,float32)、duration(时长,float32)、year(年份,int32)、hydrophone(水听器,string)。 - 标签:无标签,适用于无监督或自监督预训练。
数据集规模
| 配置 | 拆分 | 样本数 | 时长(秒) | 时长(小时) |
|---|---|---|---|---|
| default | train | 28,410 | 367,792.80 | 102.165 |
| default | validation | 3,370 | 43,629.20 | 12.119 |
| default 总计 | 31,780 | 411,422.00 | 114.284 | |
| review-sample | train | 430 | 6,017.60 | 1.672 |
| review-sample | validation | 50 | 642.40 | 0.178 |
| review-sample 总计 | 480 | 6,660.00 | 1.850 |
完整数据集覆盖情况
按年份分布:
| 年份 | train | validation | 总计 |
|---|---|---|---|
| 2019 | 920 | 119 | 1,039 |
| 2020 | 1,420 | 179 | 1,599 |
| 2021 | 14,376 | 1,820 | 16,196 |
| 2023 | 9,885 | 1,055 | 10,940 |
| 2024 | 1,809 | 197 | 2,006 |
按水听器通道分布:
| 通道 | train | validation | 总计 |
|---|---|---|---|
| channel_0 | 20,808 | 2,599 | 23,407 |
| channel_1 | 7,130 | 741 | 7,871 |
| channel_2 | 472 | 30 | 502 |
按年份与水听器通道分布:
| 拆分 | 2019 ch0 | 2020 ch0 | 2021 ch0 | 2023 ch0 | 2023 ch1 | 2023 ch2 | 2024 ch0 | 2024 ch1 |
|---|---|---|---|---|---|---|---|---|
| train | 920 | 1,420 | 14,376 | 3,982 | 5,431 | 472 | 110 | 1,699 |
| validation | 119 | 179 | 1,820 | 464 | 561 | 30 | 17 | 180 |
| 总计 | 1,039 | 1,599 | 16,196 | 4,446 | 5,992 | 502 | 127 | 1,879 |
Review Sample 覆盖情况
| 拆分 | 年份分布 | 水听器通道分布 |
|---|---|---|
| train | 2019: 12; 2020: 33; 2021: 218; 2023: 137; 2024: 30 | channel_0: 336; channel_1: 93; channel_2: 1 |
| validation | 2021: 20; 2023: 30 | channel_0: 37; channel_1: 11; channel_2: 2 |




