- 404 -Mazzola, Guerino / Noll, Thomas / Lluis-Puebla, Emilio: Perspectives in Mathematical and Computational Music Theory 
  Erste Seite (1) Vorherige Seite (403)Nächste Seite (405) Letzte Seite (454)      Suchen  Nur aktuelle Seite durchsuchen Gesamtes Dokument durchsuchen     Aktuelle Seite drucken Hilfe 

path is a path H* with a maximal value, i.e. eval(H*) >= eval(H) for all paths H of the same length. Best paths have the property that all their partial sub-paths are best sub-paths too. This is implied by the property of order-preservation. The Viterbi algorithm for best path calculation is based on this fact and works like this: For each H1 (- HARM one runs through all H0 (- HARM , calculates eval1((H0, H1)) and stores the maximal value maxV al1(H1) among these as well as the list Predecessors1(H1) of all those H0 for which the maximal value maxV al1(H1) = eval1((H0,H1)) is obtained. According to the order preservation of locusV al1 the maximum maxV al1(H1) equals
locusVal1(H0 (- mHaAxRM(transV al1(eval0(H0),H0 ~ H1),H1))),
i.e. locusVal1 has to applied just once, namely to the maximum of the transition values towards H1 which saves calculation time. Suppose now that we have already calculated maxV alk-1(Hk-1) as well as P redecessorsk-1(Hk -1) for all Hk- 1 (- HARM . At index k we fix each element Hk (- HARM , run trough all Hk -1 (- HARM , and similarly calculate
maxV alk(Hk) = locusV alk( maxHk -1 (- HARM (transV alk(maxV alk-1(Hk-1),Hk-1 ~ Hk),Hk))),
an collect those precedessors Hk- 1 which actually yield this maximal value in the set P redecessorsk(Hk) .

Best Paths are obtained backwards, starting from a locus Hn (- HARM for which maxV aln(Hn) > 0 is maximal compared to all other H'n (- HARM . Given such a best final locus Hn one selects a locus Hn-1 from Predecessorsn(Hn) , a locus Hn -2 from P redecessorsn-1(Hn -1) and so forth until a full path is selected backwards. It is obvious from the above construction that one obtains all best paths by browsing through the implied graph of possible predecessors in the described way.

1.5 Formulas for transVal and locusV al

In the present subsection we provide specific formulas for locusValk and transValk in accordance with the concrete music-theoretical examples to be discussed in the following sections. Suppose we are given

  1. a Riemann Logic RL : CHORDS × HARM --> [0, oo ) and
  2. a Harmonic Transition Value Map HT : HARM × HARM --> [0, oo )

A positive constant c > 0 regulates the relative influence of the transition values against the locus values. In the unmarked case we choose c = 1 , while c > 1 gives higher weight to transitions and c < 1 gives higher weight to local significations.

Further we suppose that we are given a chord sequence S = (Xk)k=0,...,n and finally, a sequence of custom restriction maps rk : HARM --> [0,1], (k = 0,...,n) .


Erste Seite (1) Vorherige Seite (403)Nächste Seite (405) Letzte Seite (454)      Suchen  Nur aktuelle Seite durchsuchen Gesamtes Dokument durchsuchen     Aktuelle Seite drucken Hilfe 
- 404 -Mazzola, Guerino / Noll, Thomas / Lluis-Puebla, Emilio: Perspectives in Mathematical and Computational Music Theory