Continuous-event Neural Data structure (CND)

In brief (aimed at the participants to CNSP2022)

A simple standardised data format defined to encourage standardisation, replicability, and reusability of the code. The CND format will make your life easier during and after this workshop. Let us give you three reasons to use the CND data-format:

  • By converting your data to CND, you will be able to run the CNSP-Workshop analysis scripts without changes. While this will be very useful for the mini-project, it is also an opportunity to make your future code reusable across projects;
  • This will facilitate the comparison of a particular analysis procedure on different datasets (e.g., envelope TRFs from the music and speech datasets just by changing one line of code) and, vice versa, to compare different analysis pipelines on the same dataset;
  • Standardisation will facilitate collaboration between research teams and, in the context of the CNSP-Workshop, it will allow us to answer your questions more rapidly and effectively.
We think it is important to keep things simple and to keep being aware of what exactly your code is doing. To this end, we did our best to avoid turning this work into a black-box and, instead, opted for a simple standardisation of the data-format that will make your work easier and faster while allowing you to have full understanding and control over your scripts.

CND Data Structure Specifications

The CNSP Initiative
The CNSP initiative aims to develop and collect resources, such as analysis scripts and publicly available neural data, for the study of cognition and natural sensory perception. In doing so, we propose a standardised pipeline for recording, analysing, storing, sharing, and comparing datasets on sensory perception involving naturalistic tasks, such as listening to speech and watching a movie. In addition to featuring young researchers at the top of their respective fields of research and connecting scientists from a variety of disciplines (e.g., linguistics, psychology, computer science, engineering), the CNSP workshops provide guidelines and standardised practical and educational resources for analysing continuous-event neural data. Please visit our website at and stay tuned!

What this document is about
Please find below the specifications for the Continuous-events Neural Data structure (CND). This document describes how a CND dataset should be organised, both in terms of folder structure and data structure. The guidelines can be then used to store your own data or to convert publicly available data for then availing of the CNSP resources with minimal or no changes to the analysis pipeline. These specifications will be regularly maintained at the link What kinds of experiment designs are typically considered in the CNSP resources The typical scenario considered here consists of neural signals (e.g., EEG, MEG) recorded as participants performed a natural listening task (e.g., speech listening). We will typically refer to this dataset, which is publicly available and whose CND data structure is available on the CNSP resources webpage. Other types of experiments that involve continuous sensory stimuli (music, artificial sounds) and other response measures (pupillometry, heart rate) are also possible. Interestingly, the CND data structure is also compatible with experiments involving various other continuous tasks, for example continuous motor movements. If you are planning to convert your own EEG/MEG/iEEG data, you can also refer to the BYOData preparation document. Please feel free to contact us, if you have any questions.

What is different between CND and existing data structures?
There exist standardised data structures that allow us to store a large variety of datasets from many technologies and with any experimental paradigm. However, the available solutions are either technology specific (e.g., formats for saving raw data) or general purpose (e.g., BIDS). The CND data structure is a step closer to data analysis, as it was designed to be immediately compatible with toolboxes in the area of natural sensory perception, such as the mTRF-Toolbox and the Eelbrain Toolkit. As such, the CND data structure works at a different layer than general purpose structures such as BIDS, as it is domain-specific (CNSP domain) and technology independent. Indeed, it is our intention to provide conversion scripts between CND and the most common general purpose data structures, providing the community with a rapid way to analyse and compare the increasingly large (yet heterogeneous) set of publicly available neural data in the domain of natural sensory processing.

The CND data structure
Please refer to this document for the detailed CND specifications.