AudioTag.Info - Music Recognition Robot
Reply #22 – 2009-04-11 21:26:45
null-null-pi, in general, there are no miracles. Transducing via air generally causes very serious damage to the audio that reduces its recognition chances significantly. I can only say that with GSM recordings the algorithm managed mostly well. However, it would be interesting to get/analyze your recordings used for tests (those you obtained by recording from your headset). Can you please put them somewhere for me? Addition: in general, criticism is very good but when you say that you're not sure in the robustness of this algorithm, can you please point me on any other service (free or commercial) available on the net which would recognize your test samples? unfortunately i overwrote the recorded files out of sheer lazyness (untitled.mp3 was defaulted, so i just clicked "save" every time. didn't think i would still need them). but i can re-record the samples if you like. i have to confess i didn't test ANY web based recognizer until today, so indeed i can't compare directly. but i once tested "TrackID" on my sonyericsson cellphone in a totally overcrowded disco and it identified the music at first attempt. that's why i was expecting this to work out, maybe i was a little bit overoptimistic. and sorry for my unprecise language, i'm not a native speaker so maybe sometimes i will choose wrong words and it may sound more aggressive than originally intended. edit: typo again (and again) edit3: not unnamed.mp3 but untitled.mp3