• unstressed (shorter, less aspirated) versions as in "minty," "Mindy," "moulted," and "molded,"
  
  
• voiced flaps as in "rater" and "raider,"
  
  
• glottalized or nasal released stops, as in "sweeten" and "Sweden,"
  
  
• deleted allophones, as sometimes occur in "pentagon."
  
  

Morpheme structure can be important in determining the acoustic realization of consonants. For example, /p,t,k/ are not strongly aspirated in /sp,st,sk/ clusters, except for the case where there is an obvious morpheme boundary after the /s/, as in, e.g., "discourteous" and "miscalculate" (Davidsen-Nielsen, 1974). The morpheme boundary symbol must be present in the abstract linguistic description for such words if the aspiration feature is to be computed correctly. Otherwise, a principle of assigning the maximum number of prevocalic consonants to a medial stressed vowel, subject to the constraint that the consonants form a legitimate word-initial consonant cluster (Hoard, 1966, 1971), will group the /s/-plosive into a prestressed cluster. This syllabification principle is used in Klattalk, resulting in reduced aspiration for [p,t,k] in words like "discourteous" unless a morpheme boundary is inserted after the /s/.

Prevocalic and postvocalic allophones may differ in acoustic aspects related to the temporal buildup/ decay of the sound source. Coker and Umeda (1975) observed that the prevoicing for [b,d,g] is weaker and less rich in higher harmonics in utterance-initial positions due to the more sinusoidal nature of vocal fold vibrations at initiation of voicing. Similarly, [m,n,l] were a few dB weaker in intensity (during the early portion of the consonant) in word-initial positions than in medial and final positions. On the other hand, the noise intensity for [s] was about 3 dB more intense word initially than medially and finally, presumably due to the slightly higher subglottal pressure (or the timing or pressure buildup/decay) associated with initial versus utterance-final consonants (Umeda and Coker, 1974).

In a search for sentence-level recoding rules, Oshika et al. (1975) noted the palatalization of word-final alveolar consonants if the next word begins with a palatal consonant, as in "did you" and "this shoe" . Zue and Shattuck-Hufnagel (1979) found the effect to be asymmetrical, applying to the [s] in "this shoe" but not to either the [s] or the of "wish some."

Broad and Fertig (1970) examined a collection of about 150 different nonsense words spoken by a single trained speaker. They measured formant values at ten equally spaced locations throughout each syllable, and then performed averaging over time and tokens to obtain formant values associated with the vowel. Next, they measured an average formant transition for each initial consonant, Ci, averaged over all possible final consonants, Cf. They represented this transition as a difference between the measured trajectory and the average formant position for . They observed that formant transitions associated with plosives were generally restricted to about half the vowel duration, as shown at the top in Fig. 28, but sonorant consonants often affected the entire vowel. They tried to determine whether average formant transitions for each initial C and each final C were sufficiently regular that one could predict in detail the whole formant pattern for each individual syllable from a sum of the average trajectory and the superimposed incremental trajectories for the initial and final consonants, as