知识图谱>> field-recording-db>> 返回
项目作者： matijama

项目描述 ：
Database of annotated field recording samples that can be used for training audio labelling algorithms
高级语言：
项目地址: git://github.com/matijama/field-recording-db.git
创建时间： 2018-12-20T11:47:50Z
项目社区：https://github.com/matijama/field-recording-db
开源协议：
下载


The SeFiRe field recording dataset

The repository contains the SeFiRe dataset - a dataset of annotated field recording samples that can be used for training audio labelling algorithms. The dataset contains 5 second long clips extracted from a variety of field recordings. Each clip is labelled with one label describing its contents as:

  • s: speech
  • 1: solo singing (one voice)
  • 2: choir singing (more than one voice, can also be unison)
  • i: instrumental (vocals may also be present)
  • x: (background) noise or a very noisy clip belonging to one of the above categories.

You will find two folders in the repository. Samples contains five second audio samples, their annotations are given in sample labels.txt, which is a tab-delimited text file. External samples contains annotations of five second clips extracted from a number of online sources. We can’t include the clips in the dataset, but we provide links to audio files with the location of each extracted clip and its label in external sources.txt.

Feel free to use the dataset for your research, if you use it, please cite the following paper: