Double Bass and Probability: Hiromasa Fujihara
The borderline between music and speech processing is Hiromasa’s scientific home. As is customary in speech processing, he has a very pragmatic approach to probability theory, which sounds a bit like me, but I think Hiromasa is actually much better at probability than me. He’s done some cool stuff on simultaneous estimation segregation of the singing voice and the music accompaniment, taking into account vowels and the fundamental frequency of the singing. Also, he’s got some simpler methods that detect when the singer in a song is actually singing. This is called vocal activity detection, and I say “simple”, but it is much more difficult in music than in speech. You can find publications and stuff on his AIST website here.
On that website, Hiromasa looks a bit boring (sorry), but I guess you can’t have exciting photos on official Japanese websites. In real life, he’s actually quite unboring. And a great guitarist and bass player: he actually owns an electric double bass. And much other music-making equipment I’m quite jealous of.
Hiromasa is also the person in charge of … me! He’s the one I turn to when I need to register stuff on the Japanese websites of AIST or have to do other annoying things in Japanese. Luckily, Hiromasa really is helpful, he gets things done. This is also true when actually collaborating with him: he’s a very quick coder and when I need a new feature in one of his programs, I usually get a binary within a second. Well, very quickly anyway. Since our work is related, we have worked together quite a bit, especially on lyrics-to-audio alignment.