Quenya only preserved voiced stops after nasals and (sometimes) liquids, and in all other combinations these sounds changed. Probably the earliest such change was that pairs of voiced stops were unvoiced and ultimately became voiceless stops or stop/spirant pairs. Tolkien mentioned this change in both the Outline of Phonetic Development (OP1) from the 1940s and the Outline of Phonology (OP2) from the 1950s, first discussing them in terms of geminate pairs of identical voiced stops:

dd, bb, gg at an older stage had frequently become or been replaced by nd, mb, ñg. Where they remained unnasalized, or where reproduced, they became unvoiced > tt, pp, kk (OP1: PE19/40).

The long dynamic voiced stops: bb, dd, gg. These had been largely replaced by the nasalized forms mb, nd, ñg. Where they remained unnasalized or were reproduced, they became unvoiced > tt, pp, kk (OP2: PE19/83).

As indicated above, geminate voiced stops were rare in Primitive Elvish, since they were generally replaced by nasalized voiced stops instead (PE18/41, 91). This was not really a phonological change: the major morphological mechanism that produced geminate stops in Primitive Elvish (strengthening) preferred nasalized voiced stops over geminates. Tolkien only gave one example of a geminate voiced stop in Primitive Elvish, probably invented just to demonstrate this Quenya sound change: lubbŭ > luppo “a clumsy piece or lump” (OP1: PE19/45, OP2: PE19/92).

As a result, combinations of voiced stops mostly arose from suffixion and (more rarely) in compounds. Since b, g did not appear in suffixal elements, the most common combinations were bd, d+d, gd. However, in Primitive Elvish, [t+t], [d+d] from suffixion became [st], [zd], so that d+d > zd before the Ancient Quenya period. Tolkien thus only discussed the development of the most common remaining combinations, bd, gd, but presumably other (rarer) pairs developed similarly:

With voiced stops: only bd, gd, d-d were produced by suffixion. These became ƀd, ʒd, zd (the last in CE). They were unvoiced in AQ producing pþ [?ꝑþ], hþ, st > pþ (ps), ht, st. Since these forms were rare, and usually grammatically isolated, they survived without alteration, if at all (OP1: PE19/45).

In these gd, bd, zd, together with the “dynamic” long bb, dd, gg (see below), were intensified, before the opening of the voiced stops began, with the result that they became kt, pt, st and had the same later treatment as original kt, pt, st (OP2: PE19/92).

The exact phonetic developments were somewhat different in the 1930s and 1950s, but the end result was the same: the phonetic developments of bd, gd merged with those pt, kt. The ultimate result in the 1950s was ꝑt [ɸt], ht [xt] because ultimately [pt], [kt] became [ɸt], [xt] (OP2: PE19/84), although [ɸt] was still written as pt. In the 1930s, however, bd and pt > ps (as noted above). Tolkien gave several examples in OP1 and OP2, and more examples can be found in The Etymologies of the 1930s. These examples include:

• ✶negdē > Q. nehte “honeycomb” [1930s] or “honey” [1950s] (EtyAC/NEG; OP2: PE19/91).
• ✶khagdā > Q. hahta “pile, mound” [1930s] (Ety/KHAG, OP1: PE19/45) or “fence, hedge” [1950s] (OP2: PE19/91).
• ✶labdē > Q. lapte “gluttonous eating” [1950s] (OP2: PE19/92) vs. ᴹQ. lapsa of the same meaning (OP1: PE19/45).
• ᴹ✶libda > ᴹQ. lipsa “soap” (Ety/LIB²).

The last two examples illustrate the differing results of bd in the 1930s and 1950s. As discussed above, the path taken by the sound changes in these two conceptual stages was also different: in the 1950s the combination unvoiced first, then the first stop spirantalized, whereas in the 1930s the pair was first spirantalized and then the combination was unvoiced, with one of the pair ultimately being restopped. Thus: bd > ƀd (?ƀđ) > pþ > ps but gd > ʒd (?ʒđ) > ht [xt]. This means the unvoicing of the pair was before voiced stops became spirants in the 1950s, but after spirantalization in the 1930s with the exception of geminate pairs, which must have unvoiced first.

Conceptual Development: The only example of this sound change in the 1910s and 1920s is ᴱ✶dagd- > ᴱQ. laira- “to wear out” vs. ᴱT. daida- (PE14/66). In this example, it seems the pair spirantalized and then the ʒ vocalized, similar to the phonetic development of voiced stops before liquids.

The conceptual developments between the 1930s and 1950s are discussed above. The reordering of the unvoicing vs. spirantalization sound changes barely matters, but the revised development of (pt > ps) in the 1930s vs. (pt > ꝑt [ɸt]) in the 1950s does make a difference. For simplicity, the Eldamo data model represents this phonetic development with unvoicing occurring first in both the Middle and Late conceptual periods, to avoid splitting this sound change up into two separate rules for geminate and non-geminate pairs.