proposal.tex

   1 %&proposal
   2 \begin{document}
   3 \maketitle
   4
   5 My proposed research consists of two questions of which the thesis will answer
   6 at least one.
   7
   8 The first topic is singing voice detection. Singing voice detection has been
   9 done on numerous amounts of musical styles ranging from unconventional styles
  10 like Beijing opera to conventional pop music. Moreover, the problem has been
  11 tackled using myriads of different approaches such as HMMs with different
  12 acoustic model types but also machine learned feature sets. I would like to
  13 explore how HMM based techniques perform on extreme heavy metal styles to see
  14 how well it can detect growling and how classifier might be adapted to perform
  15 better. Initially the classifier will be a binary classifier that classifies
  16 growling and non-growling. Later on classes might be added such as and musical
  17 genres.
  18
  19 Singing voice detection is often used as a preprocessing step for song lyrics
  20 forced alignment. If the time permits I would like to explore forced alignment
  21 using existing phone models on extreme heavy metal styles. Features probably
  22 need to be changed to improve performance since growling is very different from
  23 regular singing and speaking.
  24
  25 The data for this will be coming from my personal collection audio CDs. For the
  26 singing voice detection the data of one band can be used and even a test set
  27 can be held out since the band made over 15 full studio albums each with a
  28 running time between 30 and 60 minutes. The segmentation for the trainingsdata
  29 can be done by hand. Later, for the lyrics alignment, the labels for the
  30 segments can be found online on song lyric websites. Validation of the
  31 alignment is a bit tricky however since there is no golden standard but my own.
  32
  33 \paragraph{Planning}\strut\\
  34 This results in the following rough outline divided on a month by month basis
  35 shown in Table~\ref{tbl:outline}.
  36 Possible pitfalls can arise in preparing the data since that requires
  37 segmentation. It is expected to take around twice the playing time but that
  38 might be an overestimation.
  39
  40 \begin{table}[ht]
  41         \centering
  42         \begin{tabular}{cll}
  43                 \toprule
  44                 Month & Description\\
  45                 \midrule
  46                 March
  47                         & Preparing the data\\
  48                         & Preparing an experiment platform\\
  49                         & Literature research\\
  50                 April
  51                         & Running the experiments\\
  52                         & Fiddle with parameters\\
  53                         & Explore the possibilities for forced alignment\\
  54                 May
  55                         & Write up the thesis\\
  56                         & Possibly do forced alignment\\
  57                 June
  58                         & Finish up thesis\\
  59                         & Wrap up\\
  60                 \bottomrule
  61         \end{tabular}
  62         \caption{Outline}\label{tbl:outline}
  63 \end{table}
  64
  65 \end{document}