• Complete coverage with syllables

  • -- cannot model all syllables (lack of training data)

    -- in current approach, every syllable is a syllable model or all phones

    -- DOG => _d_ao_g

    -- DOGS => d ao g z

    -- possible solution is to use phones with syllables to model words with suffixes

  • Ambisyllabic consonants

  • -- current approach is to treat ambisyllabics specially in syllables but not in phones

    SHADING => sh ey d _#d_ih_ng