Systems for user and application mobility in wired and wireless networks

The feature extractor is the most relevant block, since it greatly accounts for the overall effectiveness of the system. On the other hand, a highly effcient seeker must be implemented (see 2.4).
We recall that the feature extractor is an algorithm that must produce the fingerprint of a piece of audio, by selecting some of its distinguishing characteristics.
Among the many methods investigated in the last years, we selected that proposed by Philips in 2001, in its revised and extended version .
Beyond its experimented effectiveness, we mainly based our choice on its great flexibility. According to its original proposers, we will denote the feature extractor algorithm as Streaming Audio Fingerprinting (SAF), but it is also known in the literature as Philips Robust Hash
(PRH).
The Philips approach provides a convenient and easy-to-handle output (a bit matrix), and ample opportunity of customization. Moreover, its simple steps allow some kind of statistical modeling, though only approximated .
Let's follow Fig. 2.2. The input is constituted by raw PCM samples, whose fre- quency must be 44 100 Hz, with a quantization level of 16 bit/sample. A stereophonic input is immediately converted to monophonic by averaging the two channels.