This paper discusses important scalability issues exhibited by the high-dimensional, automatic audio recognition problem. The emphasis is especially put on a well-known, robust fingerprint algorithm. Extensive tests on a very large database show the importance of the parametrization of the algorithm, in order to afford an efficient search strategy. We also quantitatively verify the capital attention that should be devoted to the caching issue, in order to avoid disk access as much as possible. Comparisons and insights are provided also with the respect to the bruteforce approach.
File in questo prodotto:
Non ci sono file associati a questo prodotto.