CAVA

A Human Communication Audio-Visual Archive for UCL

• Home
• News
• About CAVA
• FAQ
• Documents
• Publications
• CAVA Repository
• Contacts

Documents - Project Report

File formats for use in the CAVA repository

Matt Mahon, CAVA Project Officer, July 2009 - Updated October 2010

« Back to documents

• 1. Introduction
• 2. CAVA data
• 3. CAVA file creation workflow: overview
• 4. File formats
• 5. Summary of CAVA format specifications

1. Introduction

The CAVA project will build a repository to collect and share audio-visual datasets created by UCL's Human Communication researchers. This short document summarises the file formats to be used in the CAVA repository.

2. CAVA data

The data which will be placed in the CAVA repository comes from a wide range of sources and in a wide range of formats. Consequently it has a wide range of software requirements, depending on the equipment used to make the recordings. Our aim is to introduce uniformity where practical, and maintain master copies of all the data, as well as both an audio-only and a compressed video copy for dissemination. If appropriate, we will also produce a small sample video from each data set available by streaming at the collection level, so that potential users can explore the repository and select the collections most appropriate to their work.

3. CAVA file creation workflow: overview

This diagram shows the creation of dissemination files from an uncaptured mini-DV. In the case of data that is born-digital, the first capture stage is unnecessary, and the original digital file is taken to be the master.

4. File formats

4.1 Master copies
For CAVA purposes, a container file like an AVI contains four objects: video data, audio data and video and audio codecs. The codecs provide the information about how the data are stored within the file and how to uncompress them for viewing. A program which is able to correctly identify and open a container file might not be able to subsequently decode the actual data stored within, because either the metadata in the container file are not sufficient, or the software lacks the specific codec, to interpret the actual data the file contains. So a container format is only as useful as the codec it is packaged with, and the technical aspects of the video and audio data it contains can be altered as appropriate.

The capture files from Mini-DV have a data rate of 28.8mbps, which is very large (streaming videos, for instance, tend to have a data rate of around 300kbps). These files are unwieldy, but will rarely be accessed. Master files from Mini-DV, which make up the majority, will maintain their original data rate, as will high-quality recordings from any recording device. The DV codecs used in the original files will often be proprietary codecs which are built into the hardware used to record the videos. All DV-AVIs will be converted to an industry standard DV25 codec. At this stage it is appropriate to apply a filter to remove deinterlacing artefacts.

Originals from other formats may have differing data rates, and every effort will be taken to maintain their quality. Low-quality legacy data, such as those which only exist in MPEG-1 format, will not be stored at higher resolutions or data rates. The aim is to simply maintain the data. The audio will remain uncompressed. By these means the full quality of the original, whatever that may be, will be preserved.

4.2 Dissemination: Video
The CAVA video download format must be one which is small enough to download and manipulate, and which has a widely-recognised codec. MPEG-1 video with an MPEG-1 layer II audio codec (MP2) is the most appropriate, given that the dissemination copies must work on UCL-managed computers. A data rate of around 3024kbps is preferred, in order to strike a balance between high-quality preservation copies and manipulable downloads. Where lower quality MPEG-1s exist, as with some legacy data, a minimum data rate of 1000kbps is acceptable. These dissemination files are suitable for analysis using analytical software such as DHCS Videolab, CLAN (CHILDES), Transana, Elan and Praat, among others.

When data exists in other formats (MPEG-4, MOV), it will not be converted to MPEG-1. Born-digital recordings may have been made in high quality or high definition (HD). In these instances compressed versions with a similar data rate (~3000kbps) should be supplied by the depositor.

4.3 Dissemination: Audio-only
The audio-only files to be produced by CAVA are uncompressed (PCM), in WAV or AAC format, 44100Hz, 16-bit stereo with a data rate of 1024kbps, which is industry standard CD quality. Given the relatively small data rate of high-quality audio, there is no reason not to maintain the highest quality possible here. Note that audio data from natural conversation may occasionally be unsuitable for speech processing software, where it was not recorded using appropriate hardware or formats.

4.4 Dissemination: Streaming
The UCL video streaming server uses files in Flash Video format (FLV), with an ON2 VP6 video codec. The data rate is 400kbps. This is an aggressive compression, but one which does not significantly reduce the subjective quality of the data. Both the download video and the streaming video use lossy compressed audio. As MP2 and MP3 are standards for the file types, there is little that can be done to change this. The sample videos are intended to give the user a taster of the content, rather than the quality, of the datasets.

The DigiTool platform that CAVA uses will attempt to stream higher-quality videos when they are accessed in Internet Explorer. CAVA recommends using Firefox. Please see the Troubleshooting pages for advice on how to configure your browser for CAVA.

5. Summary of CAVA format specifications

When data exists in other dissemination-quality formats (MPEG-4, MOV), it will not be converted to MPEG-1. Born-digital recordings may have been made in high quality or high definition (HD). In these instances compressed versions with a similar data rate (~3000kbps) should be supplied by the depositor.

« Back to documents