Among the Setouchi dialects of the Amami region of Northern Ryukyuan (a close relative of Japanese) -- I'm now speaking of Shodon dialect specifically -- voiceless obstruents in initial position are either aspirated or not, and the distinction cannot always be predicted from the following vowel. In addition, there is a set of voiced obstruents. All the obstruent manner distinctions are lost in final position, typically through morphophonemic processes of (historical-)high-vowel loss. (Strictly speaking, the aspiration distinction does not obtain in the syllable where the vowel can be dropped in any case, since originally aspiration occurred only before non-high vowels, and the apocope occurs only with high vowels. I don't know, then, if this case fits your criteria. There IS a neutralization of voicing between obstruent pairs in this position.)

Leon A Serafim

formerly Associate Professor of Japanese,

University of Hawai'i at Mānoa

currently retired and living in Santa Cruz, California

In Fulfulde also spoken in Guinea Conakry in West Africa, the phonetic feature of "alternance" (in French) exists. According to this feature, whether or not a voiced consonant maintains or loses its voice (neutralization) is susceptible to be a function of the syllabic position. Typically, three syllabic positions can be distinguished, namely initial, mid and final. The voiced initial consonant can lose its voice and become voiceless in final position. Refer to studies by Labatut and others.

In Dutch we have consonants that are distinguished, among other things, in voiceless and voiced consonants. The vocal cords vibrate with the formation of voiced consonants; in the formation of voiceless consonants not. Dutch has the following voiceless consonants: p, t, k, f, s, ch and h, and the following voiced consonants: b, d, v, z, g, m, n, ng, l, r, j and w.

The position doesn't matter because the voiceless consonants are always voiceless.

Don't know if this answers your question?

HI, Scottish Gaelic has final obstruent devoicing. E.g. mag /mak/ 'deride', but magadh /'magəɣ/ 'derision', as against mac 'son'. However, the final obstruent in mac is pre-aspirated, giving /'maʰk/ or /maxk/, so they are distinct (but not in their devoicing). Similarly fad /fat/ 'length', cat /kʰaʰt/ 'cat', etc. Initial voiceless obstruents are all post-aspirated, e.g. /kʰaʰt/ again, /pʰi:s/ 'piece', /tʰarəf/ 'bull', ... Hope this helps.