Add AedTreeBuilder

[larissakl]

Adds an AedTreeBuilder which can later be used for a TreeLabelsyncBeamSearch. The search tree only includes labels, without blanks, self-loops and skip-transitions.
The sentence-end token is retrieved from a special lemma in the lexicon and added as a label reachable from the root.
As in the CtcTreeBuilder, a word-boundary root will be added if a word-boundary token is present in the lexicon.
I moved some helper functions from CtcTreeBuilder to AbstractTreeBuilder so that I can easily reuse them.

[larissakl]

The helper functions used by CtcTreeBuilder and AedTreeBuilder are now in the shared base class CtcAedSharedBaseClassTreeBuilder SharedBaseClassTreeBuilder instead of AbstractTreeBuilder.

[SimBe195]

@larissakl Do you have any plots that show the generated tree structure for a simple example lexicon? If so, it would be nice to include one in this PR.

[larissakl]

Sure, here is an example tree and this is the corresponding example lexicon:

<?xml version="1.0" ?>
<lexicon>
  <phoneme-inventory>
    <phoneme>
      <symbol>_</symbol>
      <variation>none</variation>
    </phoneme>
    <phoneme>
      <symbol>A</symbol>
      <variation>none</variation>
    </phoneme>
    <phoneme>
      <symbol>B</symbol>
      <variation>none</variation>
    </phoneme>
    <phoneme>
      <symbol>C</symbol>
      <variation>none</variation>
    </phoneme>
    <phoneme>
      <symbol>[SILENCE]</symbol>
      <variation>none</variation>
    </phoneme>
    <phoneme>
      <symbol>[UNKNOWN]</symbol>
      <variation>none</variation>
    </phoneme>
    <phoneme>
      <symbol>&lt;/s&gt;</symbol>
      <variation>none</variation>
    </phoneme>
    <phoneme>
      <symbol>@</symbol>
      <variation>none</variation>
    </phoneme>
  </phoneme-inventory>
  <lemma special="silence">
    <orth>[SILENCE]</orth>
    <orth/>
    <phon>[SILENCE]</phon>
    <synt/>
    <eval/>
  </lemma>
  <lemma special="unknown">
    <orth>[UNKNOWN]</orth>
    <phon>[UNKNOWN]</phon>
    <synt>
      <tok>&lt;UNK&gt;</tok>
    </synt>
  </lemma>
  <lemma special="sentence-end">
    <orth>&lt;/s&gt;</orth>
    <phon>&lt;/s&gt;</phon>
  </lemma>
  <lemma special="word-boundary">
    <orth>@</orth>
    <phon>@</phon>
  </lemma>
  <lemma>
    <orth>AA</orth>
    <phon>A A</phon>
  </lemma>
  <lemma>
    <orth>AB</orth>
    <phon>A B</phon>
  </lemma>
  <lemma>
    <orth>AAA</orth>
    <phon>A A A</phon>
  </lemma>
  <lemma>
    <orth>AAB</orth>
    <phon>A A B</phon>
  </lemma>
  <lemma>
    <orth>ABA</orth>
    <phon>A B A</phon>
  </lemma>
  <lemma>
    <orth>ACA</orth>
    <phon>A C A</phon>
  </lemma>
  <lemma>
    <orth>BA</orth>
    <phon>B A</phon>
  </lemma>
  <lemma>
    <orth>BAC</orth>
    <phon>B A C</phon>
  </lemma>
</lexicon>

m=... is the AM index (1 for A, 2 for B, 3 for C, 4 for [SILENCE], 5 for [UNKNOWN] and 6 for sentence-end, just as the order of the phonemes). The blank symbol _ is still part of the lexicon, but not relevant for this tree anymore.

[SimBe195]

It looks like "word-boundary" is part of the lexicon but not in the tree. From the code it looks like word-boundary should be integrated though. Is this really the right picture given the lexicon?

[larissakl]

Oh yes, you're right. This was the tree without word-boundary token. With word-boundary in the lexicon, it looks like this: