Eloi Batlle

A Review of Audio Fingerprinting

An audio fingerprint is a compact content-based signature that summarizes an audio recording. Audio Fingerprinting technologies have attracted attention since they allow the identification of audio independently of its format and without the need of meta-data or watermark embedding. Other uses of fingerprinting include: integrity verification, watermark support and content-based audio retrieval. The different approaches to fingerprinting have been described with different rationales and terminology: Pattern matching, Multimedia (Music) Information Retrieval or Cryptography (Robust Hashing). In this paper, we review different techniques describing its functional blocks as parts of a common, unified framework.

Audio Watermarking and Fingerprinting: For Which Applications?

Although not a new issue, music piracy has acquired a new status in the digital era, as recordings can be easily copied and distributed. Watermarking has been proposed as a solution to this problem. It consists in embedding into the audio signal an inaudible mark containing copyright information. A different approach, called fingerprinting, consists in extracting a “fingerprint” from the audio signal. In association with a database, this fingerprint can be used to identify a recording, which is useful, for example, to monitor audio excerpts played by broadcasters and webcasters. There are far more applications to watermarking and fingerprinting. After a brief technical review, this article describes potential applications of both methodologies, showing which one is more suitable for each application.

Content-based identification of audio titles on the Internet

The increasing usage of the Internet for the distribution of audio content and the increasing number of audio broadcasting stations require new supporting tools and methods for the observation of occurrence frequencies and of possible copyright infringements. The paper describes a general approach for the identification of audio titles and its application on Internet observation. The concept of an AudioDNA is developed, allowing a highly compressed representation of a sequence of acoustic events. Audio titles are identified by using a sequence matching method which determines similarities between observed and reference AudioDNA stored in a database. This method is implemented in a highly scaleable architecture allowing the identification of several audio titles in parallel. The first results with this system are very promising, only highly distorted audio titles are not identified correctly.

Acquisition of semantic classes for adjectives from distributional evidence

In this paper, we present a clustering experiment directed at the acquisition of semantic classes for adjectives in Catalan, using only shallow distributional features.We define a broad-coverage classification for adjectives based on Ontological Semantics. We classify along two parameters (number of arguments and ontological kind of denotation), achieving reliable agreement results among human judges. The clustering procedure achieves a comparable agreement score for one of the parameters, and a little lower for the other.

Recognition and analysis of audio for copyright protection: the RAA project

Automatic generation of play lists for commercial broadcast radio stations has become a major research topic. Audio identification systems have been around for a while, and they show good performance for clean audio files. However, songs transmitted by commercial radio stations are highly distorted to cause greater impact on the casual listener. This impact helps increase the probability that the listener will stay tuned in, but the price we have to pay is a severe modification in the audio itself. This causes the failure of traditional identification systems. Another problem is the fact that songs are never played from the beginning to the end. Actually, they are put on the air several seconds after their real beginning and almost always under the voice of a speaker. The same thing happens at the end. In this article, we present the RAA project, which was conceived to deal with real broadcast audio problems. The idea behind this project is to extract automatically an audio fingerprint (the so-called AudioDNA) that identifies the fragment of audio. This AudioDNA has to be robust enough to appear almost the same under several degrees of distortion. Once this AudioDNA is extracted from the broadcast audio, a matching algorithm is able to find its fragments inside a database. With this approach, the system can find not only a whole song but also small fragments of it, even with high distortion caused by broadcast (and DJ) manipulations.

AMADEUS: a scalable HMM-based audio information retrieval system

The new transmission and storage technologies now available have put together a vast amount of digital audio. All this audio is ready and easy to transfer but it might be useless with a clear knowledge of its content as metadata attached to it. This knowledge can be manually added but this is not feasible for millions of on-line files. In this paper we present a method to automatically derive acoustic information about audio files and a technology to classify and retrieve audio examples.

Scalability issues in an HMM-based audio fingerprinting

Audio fingerprinting technologies allow the identification of audio content without the need of external meta-data or watermark embedding. These audio fingerprinting technologies work by extracting a content-based compact digest that summarizes a recording and comparing them with a previously extracted fingerprint database. In this paper we present a fingerprint scheme that is based on hidden Markov models. This approach achieves a high compaction of the audio signal by exploiting structural redundancies on music and robustness to distortions thanks to the stochastic modeling. In this paper we present the basic functionality of the system as well as some results

Visualization of metre and other rhythm features

We present a new rhythm, metre and BPM visualization tool. It is based on the proposed rhythm transformation that transforms audio data from time domain to a so-called rhythm domain. The goal of this method is that data in rhythm domain can be interpreted as frequency domain information (for BPM detection) as well as time domain information (for metre detection). Some musical information, such as the metre (simple or compound, duple or triple, swinged or non-swinged), can be extracted from input audio. The method is based on the periodogram computation of the processed input data, and the different musical features are extracted by using well known techniques.