It seems that in Old Sindarin and Old Noldorin, a final [-d] after a vowel became a spirant [-ð] and then vanished with compensatory lengthening. This sound change did not occur when the [d] was protected by a final vowel that was lost later (where it developed instead to [-ð]), and the phonetic development may have been limited to monosyllables. We only have a single example of this sound change, but it appeared multiple times in Tolkien’s writing, for both Sindarin and Noldorin:

• ᴹ√TA > ON. tō “thither” (Ety/TA).
• ᴹ✶tā̆d > ON. tō “thither” (PE21/58).
• ᴹ✶tad > tā > N. tō “thither” (PE19/52).
• ✶tad > tað > tā > S. taw² “thither” (PE19/104).

The primitive form of this word was formed by combining the demonstrative root √TA with the ancient allative suffix -d(a), remnants of which can also be seen in its Quenya cognate tar. The Sindarin example provides the most complete etymology, which seems to be [tad] > [tað] > [tā] > [tǭ] > [tau]. The Sindarin form has two lenited variants: daw and do, the latter probably with shortened [ǭ] > [o] in unstressed positions.

Another possible example of this sound change appears in a chart of Quenya and Sindarin numbers from the late 1950s or early 1960s (PE17/95), which shows:

					lepen	enk-
Q.	er, min	atta	nel-de	can-ta	lempe	enque
S.	min, er	tad	nel(eð	can(ad	leb(en	eneg-

The number nel(eð might represent a variant number nel resulting from the loss of final [-d] from the root √NELED, as can(ad and leb(en might represent loss of final [-t] and [-n] from √KANAT and √LEPEN. In this case, however, other explanations are possible, such as derivation from reduced forms of the roots: √NEL, √KAN, √LEP.

There isn’t enough information to determine whether similar changes would happen with other final voiced stops [-g] or [-b]. The former would probably become [-ɣ] and vanish with compensatory lengthening, while the latter probably became [-β] > [-u]. These theoretical sound changes are hard to detect, since unlike [d], similar developments would have occurred in cases where primitive [g] or [b] was followed by a final vowel that was later lost.

Determining the timing of this change is hard, given the lack of examples. However, since the lengthened vowel [ā] followed the same development as ancient vowels, it was probably quite early in Old Sindarin/Old Noldorin, and thus occurred well before the general change whereby voiced stops became spirants after vowels.

See OS. final [d] spirantalized and vanished for discussion.