Back RAA


Recognition and Analysis of Audio

Reference: FP5-IST-1999-12585
Duration: from 01/02/2000 to 31/07/2002

Project description

Each individual audio-title is supposed to get a unique, individual AudioDNA, but not all what differs is already a different title (e.g.: audio-title on CD differs a lot from audio-title received in FM-transmitter).

In this sense a strategy has to be defined how AudioDNA should be affected by changes of the underlying audio-title:

  • identical titles

Two titles are identical when they derive from the same recording and have the same history. Typically they also have the same unique-id in a customer's database (e.g.: GEMA number).

  • identical titles affected by transmission and other distortions

In a "real" world there is always noise. Transmission and signal preparation leads to significant signal variations. Those variations are usually at a level which is not very sensitive for the human ear, thus the non-educated human is likely to hear the same. It is very important that the AudioDNA stays invariant if the average human-ear/brain does not recognise invariance. Strong distortions which lead to a significant change in human reception could also lead to different AudioDNA, but the similarity between AudioDNA of non-changed signal and altered signal should be proportional to the level of distortion.

RAA only deals with the above mentioned transformations. But there are some other aspects which are interesting but out of scope of RAA. Anyhow they might be considered in a follow-up effort:

  • same audio title played by the same artist but from different recordings.

It is absolutely necessary that the AudioDNA differentiates between such titles. For economical reasons it would be a very interesting feature to have a special match of audio-titles even if they are from different recordings and only one recording is registered in the database. Especially the needs of film-music could be covered with this feature. An example of this would be the recognition of a live recording from the AudioDNA extracted from a studio recording of the same title.

  • same title played by different artists.

This would make much sense for statistical information services. Especially for thematic analysis of music this feature could significantly reduce the number of titles to be registered before the observation makes sense. An example would be the recognition of a cover version from the AudioDNA created from the original title

  • recognition of sampled material.

Sampling is a widely used technique in contemporary dancing music, copy-right issues concerning unauthorised re-usage of titles could be cleared at this level. A coherence between the AudioDNA´s should be made possible.