Home Reports Information Technology Content

‘Audio DNA’ ensures sound approach to classification

next article

08.02.2006

Until E! 2668 AUDIOCLAS, there was no method to automatically classify audio and musical sound effects. This EUREKA project has resulted in an objective sound-effect classification system that should provide a major boost to European film, video and audio production. The software system makes it possible to speed access to major sound-effect libraries and simplify synthesis of new or combined sound effects from the stored data. Audio DNA is used to identify sounds similar in nature, such as door slams. This classification and taxonomy of sounds is an innovative new approach that is already being used to provide web access to a range of commercial sound-effect libraries for both professional and domestic use in Europe.

The AUDIOCLAS project set out to establish an ‘audio DNA’ classification system, based on the decomposition of a sound effect into several thousand finite elements. With sound effects playing a key role in film, video and audio productions, many film companies and post-production houses rely on sound-effect libraries to avoid the expense of creating specific individual sound effects. And a growing number of home video makers are discovering the creative possibilities of using such libraries too. Close co-operation between a leading UK post-production facility and a Spanish university audiovisual studies department led to the development of a fully automated approach to sound categorisation. AUDIOCLAS resulted in a software-based tool that makes it possible to catalogue sounds quickly, logically and automatically. Indeed, sound effects held within the library have already been used in the two US-produced Shrek films. Searches can now be carried out using key words or by playing a sound and asking the system to find similar effects.

Finding the way around

Having the sound-effect library is not enough. Existing sounds in such libraries are often digitised and classified manually, a problem faced by the Tape Gallery post-production house when it acquired ten million items, dating as far back as the 1940s in some cases, along with an arduous mechanical card file index system. So finding an efficient way around these huge quantities of effects was essential. As computerising the whole library using manual classification could have taken two people up to 20 years — a more effective alternative needed to be found.

The company sought technical assistance to develop an automated system through the EUREKA AUDIOCLAS project with the help of veteran EUREKA projects consultant Peter Stansfield of Wavecrest Systems. “It is fairly easy to organise lists of names and addresses using a computer,” points out Stansfield. “But what do you do about sounds? You can sort on length of clip but how do you tell which are the bangs, which are the squeaks?” The key advance in AUDIOCLAS is not matching sounds exactly, but using the audio DNA to identify similar sounds such as all the door slams. It is this classification and taxonomy of sounds that is the real innovation.

A global first

“Tape Gallery is already using the system to speed access to sound effects for its own use. And the results are being built into a new web system for commercial sound-effects supplier Sound Effects Library, enabling sounds to be previewed and selected directly on line,” says Stansfield. The tool will also be licensed to other libraries, once it has been demonstrated fully.

“This is a global first – US systems depend on interpretation of a range of emotive key words that are difficult to translate linguistically,” he claims. This audio retrieval and generation facility should also lead to an increased take-up of European facilities, meaning more film and video work coming from the USA to Europe. And the university will receive licence fees for use of its technology. “EUREKA proved to be the ideal vehicle for this project, offering a ‘low maintenance, low overhead’ approach to collaboration”, adds Stansfield.” And we benefited from access to the whole EUREKA Network.”

Catherine Shiels | Source: alphagalileo
Further information: www.eureka.be/files/:1320838

next article