In Sindarin a [z] became [ð] before the voiced stops [b] and [g], which themselves became voiced spirants [β] and [ɣ]. This sound change is distinct from the development of [z] before [d], where the [z] vanished while lengthening preceding vowel. This sound change is discussed in notes appearing in the Outline of Phonology (OP2) from the early 1950s:

In Sindarin zd > ´d, đ but zg, zb > đʒ, đƀ > đa, đu, as in nadha “fetter” [< ✶nazg-], maða “mud” [< ✶mazgō/ŭ], buðu “large fly” [< ✶buzbō] (PE19/101). [The primitive forms appear earlier on the same page in Quenya derivations.]

The words in this note are a complete list of Sindarin examples for this phonetic rule from the currently published materials. With intermediate changes added for clarity, they are:

• ✶nazg- [> naðɣ(a)] > S. nadha.
• ✶mazgō/ŭ [> maðɣ(o)] > S. maða.
• ✶buzbō [> buðβ(o)] > S. buðu.

In each case, the resulting [ɣ] or [β] became final after short final vowels vanished: ultimately these final [ɣ] became [a] and [β] became [u].

Conceptual Development: There isn’t enough information from the 1910s and 1920s for us to determine how [z] developed in combination with voiced stops in the Gnomish and Early Noldorin periods. The phonetic development of these combinations in the (Old) Noldorin of the 1930s was also different from Sindarin, aligning instead with the development of [z] before [d]. For a discussion of these Noldorin developments, see the entry on how [z] vanished before [d] lengthening preceding vowel, which also discusses the Noldorin phonetic rules: where [z] became [i] before [d], [g] and [u] before [b].

There is evidence, however, that Tolkien at least considered the Sindarin phonetic developments in the Noldorin stage. In particular, some examples from a deleted entry in The Etymologies from the 1930s seem to match the Sindarin development:

• ᴹ√MÁSAG [> madhghas] > N. maðias “softness, pliancy” (EtyAC/MASAG).
• ᴹ√MÁSAG > madhgh > N. mada “soft, pliant, yielding” (EtyAC/MASAG).

The first Noldorin example shows the medial rather than final development of [ɣ], with [ɣ] becoming [i]. The second example shows d instead of the expected ð, but perhaps this is a slip or a misreading. Both these forms were deleted when the entry was rewritten, the latter replaced by N. moe “soft, pliant” and the former without replacement.