Ahoy Ahoy!
Apsal excited sgt nih? Bkn apa, my SV wat lawatan mengejutz ke lab. Dah la masa tu baru balik dari urusan x rasmi ke town, dgn hujan lebat nya, and basah kuyup nya. Nasib la smlm dah bersungguh wat keja, so ada la benda nak present :P Tp appearance aku melucukan la, sbb seluar basah kena hujan. Nasib la ada seluar spare, seluar trackbottom utk kalo2 aku nak aerobik, so dpt la aku tukar dan tidak la aku kebekuan. Sila lihat gmbr di atas. Yg kaler grey tu sluar track aku, sememeh je dipadankan dgn cardigan kat belah atas. Baju aku kering btw, sbb pki raincoat masa kluar td.
Spt biasa, kalo jumpa dia, mmg akan ada idea yg mengujakan, even kalo dia duduk 10 minit je pon dpn kita. Aku started off ngan bgtau la, normalization part tu dah beres boss (Alhamdulillah). Tp skang ni kebuntuan sbb x tau nak guna distance measurement apa utk calculate the distance similarity, sbb aku guna real audio data and decimal numbers. Kalo nak guna City Block Distance spt yg tuan thesis buat, x seswai plak, sbb dia pakai MIDI data and binary numbers. Dah try gak convert decimal to binary numbers, tp not making much sense la pulok (ni hasil penemuan dan kebuntuan semalam).
'And without the appropriate distance measurement calculation, I would not be able to add the weights to each of these segments', kata aku dgn bersungguh2 nak tunjuk dah berusaha dah ni.
'That's a very interesting question', kata SV aku sambil menggangguk2 kan kepala tanda paham kat mana aku stuck. *Kembang kejap* (maaf, sesi poyo sedang mengambil tempat - please bear with me. Ni je la little things that make my day okeh, so let me bask in the compliment a bit longer ececeh....).
Pastu tunjuk la beberapa cara lain reseracher lain wat, yg aku baca ala2 sekali imbas je la, just to see what is out there. Ada yg plot semua point tu ke dlm graph, dan ukur beza setiap titik in graph, ada yg letak value tersendiri utk melodic contour spt (U (Up) = jika melodi nya menaik, D (Down) = jika melodinya menurun, S (Same) = jika melodinya sama). Ada plak yg bg binary number, 1 kalo naik, 0 kalo turun. Prof aku suggests bubuh threshold ke, kalo naik certain %, then tambah weight berapa berapa ke...
Aku pon dgn timidnya tanya, 'Bole ke buat suke2 method apa?'
Terus dia jawab, 'Oh yes. It is time you be creative and come up with your own ideas. This can be part of your PhD contribution.'
Aku mencelah, 'What if it's silly?' <--- nada org x dak konfiden.
SV pon jawab balik, 'To you it might be, but it might work. So, let yourself out and go crazy with ideas!'.
Laaa... gitu ke, mana la eden tahu. Selama ni, masa wat Master by Research pon, x de la suggest method sendiri. In fact, x berani nak suggest apa2, kot2 la kena hentam masa viva, like, 'You pakai method sapa ni, pandai2 wat lagu ni?', soal examiner yg menyinga. X kan la aku nak jawab, 'Sendiri bikin la, Prof'. Pengsan examiner!
Honestly tho', mmg masa MSc aku x de peluang to be creative. (and sgt x kreatif pon, even if I was forced to be one). Just followed in the footsteps of other researchers, especially bcoz my research was on traditional Malaysian music, so apa lagi, byk la guna idea yg dah org carried out with western music, but I changed the niche and dataset la kan, of course. But you gotta start somewhere, so aku x regret. At least I learned the techniques and got the results of my own, how insignificant it might be to others.
Tp tu la dia agaknya beza MSc and PhD ye. MSc ni kita nak didik self utk jd researcher; follow the right methodology, read the right stuff, get updated on the current research trend and issues. Apa buat, semua mesti mau rujuk and tunjuk mana asalnya. With PhD, ada kebebasan utk wat la method apa pon. Tp persoalannya, 'Adakah aku mampu utk berfikir secara kreatif?'
Ah... Think positive! Yang penting, usaha dan tawakkal... Insya Allah...
Ok, back to work dah. Nak perah creative juices out. Pray for me ya!
9 comments:
He he, kelakar baca pasal basah kuyup tu, sib baik la ada stand by seluar + baju ek kat office :P
So which normalisation method you end up using? Nnt PM kita la, teringin juga nak tau, kot2 boleh kait dengan MM retrieval :)
Betul! betul! Thresholding is one of the popular methods for such function. Maybe lepas ni boleh buat testing, to determine what's the 'best' threshold value for your group of data ka?
klu time viva kena tanya mcm method sapa yang u refer..pastu jawab sendiri bikin..mmg nak kena tibai klu kat mesia haha
ok aku tak tau sgt apa yang ko buat noris tapi skrang ni kat IR, ramai org guna random walk. mana tau bole kasi elok itu weight.
Aku suka baca benda2 mcm ni.. sebab dia berkenaan dgn Failure-Effort-Success. Aku masih lagi dalam stage yg 1st tu dan kau pulak nampaknya dah nak mengakhiri the 2nd stage tu.. All the best!
Audio data ko skrg dalam spectrum domain ke cepstrum domain? Cuba ko check under keyword Vector Quantization (VQ). VQ ni popular jugak utk waveform based classification. Yg aku nak highlight, dalam method ni ada module utk Distanse Measurement. Mcm2 jenis distance measurement yg org proposed utk module ni. So mungkin ni boleh senangkan pencarian DM yg sesuai utk projek ko. Just mencadang aje, aku pun tak sure appropriate ke tak.
pergh..siap sv pandai terjah tuuu...
takut le pulak.
sib baik pandai cover.
wah3... nmpk nya this post has brought all geekiness in you guys... teruja i... even my Hubs pon masa komen, he was like, woh3, takut2.... akademics have spoken.... anyway, sgt2 appreciate all the feedbacks tau... smg Allah membantu perjuangan anda jua!
mas:
x guna apa2 particular method pon. mmg 3 hari dok baca2 kan, skali tu last week, my SV masuk kjp, and I told him la kan x tau nak pilih yg mana (masa tu dah siap cari log of the pitch bagai utk match dgn western musical scale).
Pastu dia bole ckp, ooh, dia nak simple je, just normalize as in buat boundary from this pitch value to that pitch value jd one pitch value (i.e from 196 - 207 Hz jd 196 Hz sahaja, which gives the note G in the 3rd octave... and 208 - 219 Hz jd 208 Hz, note G# in 3rd octave.. you get the pic...). buat la utk all the possible pitch values for all 8 octaves...
so skang nak kena go to thresholding la kot, dah dia suggest cam tu. awak pon ckp guna thresholding mengikut value sendiri jgk ye? nnt kalo kita ada result yg dah gempaks sket, kita diskas panjang2 lg eh? sesekali diskas benda ilmiah ni rasa refreshing plak kepala otak. maklum la, sini x diskas sgt ngan sesapa....
azri:
random walking ye? itu aku slalu buat, tp pusing2 town la haha... no just kidding. to be honest, aku x pernah dgr lg method ni, so nak gi baca google la ni, apa kebenda nya itu, dah siap catit dah dlm things to do today. thanks tau. hopefully membantu la :)
izuan:
ye ke aku stage 2 dah? ntah la, kalo dah nak berjaya tu bagus la hihi...
eh, aku pki data spectrum. basic average pitch je, aku dah godak jd apa2 lg, x apply FT, STFT, DWT dsb nya lg. so kiranya lom focus on the features lg, just to get the system to work the basics dulu. style prof aku ni, dia nak system kita dah siap in the very most basic way possible, tp jalan and dpt hasil dulu, utk dipresentkan masa transfer process (after 1 to 1.5 year start phd, which is coming soon, around oct-mac for me yikes!).
bila dah ada working prototype tu, cjance utk lulus mini viva (transfer) tu adalah tinggi. kalo x lulus (nauzubillah), dpt MPhil ajo la.
so kata la lulus kan, masuk tahun 2 tu la baru reserach yg sungguh2, cari features apa, optimize paramateers itu ini, etc etc... tp working prototype dah ada patut dah bole bina from there. so far aku suka cara dia guide aku kerja, sbb aku nmpk, n aku x got stuck dlm 'complexity' of it all masa baru2 start. serius bila dah start developing baru lebih jelas nmpk apa yg kita kena wat.
spt jg azri, aku lom sempat tgk lg vector quantization tu, but i have jotted it down utk search kejap lg. berguna ke x tu secnd kira, yg penting, thanks atas idea ko itu. moga2 membantu aku....
isabelle:
tu la pasal... dia jarang betul jengah2 tanpa warning... kena plak hari kita keluar... nasib keluar kejap je, and nasibbbbb la dah ada something nak tunjuk hehe!
Post a Comment