For much of his life, Tolkien imagined an initial voiced velar spirant ʒ [ɣ] becoming breath h in Ancient Quenya and ultimately surviving in Quenya. This change must have occurred before voiced stops became spirants, otherwise we would see g- > ʒ- > h- rather than initial ʒ- from g- vanishing (which is what actually happens). Already in Primitive Elvish this ʒ-sound had vanished in most cases medially and finally, and its appearance as h- initially was one of its few survivals in Quenya. The exact development of the primitive voiced velar spirant is one that Tolkien vacillated on quite a bit throughout his life.

Conceptual Development: In the 1910s, the phonetic inventory of Primitive Elvish included both a voiced velar spirant [ɣ] and a voiceless velar spirant [x] (PE12/15). In the Qenya Lexicon [QL] of the 1910s, Tolkien represented the voiceless sound as H- in roots, and the voiced sound as Ʒ-. Unlike later Quenya, the voiceless sound survived in Early Qenya as h- but the voiced sound vanished, as opposed to Gnomish where initial ʒ- > g- as seen by examples in the contemporaneous Gnomish Lexicon [GL]:

• ᴱ√HATA > ᴱQ. hata- “hurl, fling” vs. G. hada- “throw at” (QL/39; GL/48).
• ᴱ√HELE > ᴱQ. helke “ice” vs. G. heleg (QL/39; GL/48).
• ᴱ√HONO > ᴱQ. hon “heart” vs. G. honn (QL/40, GL/49).
• ᴱ√ƷARA > ᴱQ. arda “place, spot” vs. G. gar(th) (QL/32; GL/37).
• ᴱ✶ʒalast- [marked by Tolkien with a “?”] > ᴱQ. alas “marble” vs. G. glast (QL/30; GL/39).
• *ᴱ√ƷONO [shown as ᴱ√ONO in QL] > ᴱQ. on(d) “stone” vs. G. gonn (QL/70; GL/41).
• *ᴱ√ƷUÐU [shown as ᴱ√USU/UŘU in QL, gudh in GL] > ᴱQ. usta- “to burn” vs. G. gusta- (QL/98; GL/42).

Starting with the Early Noldorin Dictionaries of the 1920s, we can see some Early Quenya examples where ʒ- > h- instead:

• ᴱ✶ʒaldá > ᴱQ. halda “wide, broad” vs. ᴱN. gall (PE13/144).
• ᴱ✶ʒirdǝ > ᴱQ. hir “inwards, interior” vs. ᴱN. gir (PE13/161).

In the 1930s, Tolkien removed the voiceless velar spirant as an independent sound from the phonetic inventory of Primitive Elvish, replacing it with the aspirate KH- [kʰ]. The roots above became: ᴱ√HATA, ᴱ√HELE, ᴱ√HONO >> ᴹ√KHAT, ᴹ√KHEL, ᴹ√KHO(N). However, he retained the voiced velar spirant ʒ in Primitive Elvish; the voiceless velar spirant [x] became an allophone of this voiced sound, appearing only before voiceless consonants (PE18/30). In The Etymologies of the 1930s, he retained the Quenya initial development of ʒ- > h- from the 1920s, but this initial sound vanished in Noldorin rather than becoming g-:

• ᴹ√ƷAN > ᴹQ. hanu “male”, N. anw (Ety/ƷAN).
• ᴹ√ƷAR > ᴹQ. harda “realm”, N. ardh (Ety/ƷAR).
• ᴹ√ƷEL > ᴹQ. helle “sky”, ON. elle (Ety/ƷEL).

Somewhat cryptically, in the contemporaneous Outline of Phonetic Development (OP1) from the 1940s, Tolkien said that initial ʒ vanished in Quenya:

ʒ, and with it ʒ derived from denasalized ñ, early disappeared (already in Common Eldarin) medially, with resultant vocalic contractions. Initially also it had vanished without trace already in Ancient Quenya (OP1, PE19/33).

The same initial-vanishing is seen in the Comparative Tables of phonetic development, also from the 1930s: ʒ > - (PE19/19). The only example from The Etymologies that I can find which demonstrates this vanishing is ᴹ√ƷOR > ᴹQ. orro “alas!”, and it appears next to a variant horro where the sound did not vanish (EtyAC/ƷOR). Such vanishings do occasionally appear in other texts from the 1930s: ʒō̆ “away from” > o, where Tolkien originally had o >> ho >> o (PE21/60). However, ʒ- > h- was the norm in The Etymologies of the 1930s.

In the second version of the Tengwesta Qenderinwa (TQ2), ʒ was still part of the phonetic inventory of Primitive Elvish (PE18/82), but Tolkien repeated his assertion that initial ʒ vanished in Quenya:

In Quenya ʒ vanishes initially without traces (TQ2, PE18/104).

Likewise, at the very beginning of his writing on the contemporaneous Outline of Phonology (OP2), he initially included ʒ as an independent sound in Primitive Elvish, but soon changed his mind, deciding that the independent sound was a weak voiceless spirant which he transcribed as h (PE19/69 note #3). It is likely that after this revision, ʒ became the voiced allophone of h (as z was for s), reversing the situation from the 1930s. At this point he decided once again that this sound was retained initially in Quenya:

In a number of correspondences Q. shows words or whole stems with initial h- where the Telerin branch all show vocalic beginning. Now this h- originally had a separate letter ½ (halla), distinct from d (product of kh), evidently representing a breath h while products of kh were still spirantal. It is probably to be explained as descending from h-breath as an independent consonant in Eldarin (PE19/74).

The H-sound continued to appear in roots up through the 1960s, though there is a sprinkling of Ʒ- thoughout, making it hard to nail down the exact timing of the revision of Ʒ >> H. Nevertheless, by the time of the Quendi and Eldar essay around 1960, H- is well established, as in the root √HO “(coming) from”, the basis of the Quenya genitive at this conceptual stage.

Sometime in the late 1960s it seems Tolkien consider restoring both voiceless and voiced velar spirants as distinct independent sounds in Primitive Elvish. In notes associated with his essay on the Variant D/L in Common Eldarin written in the late 1960s, he said:

This variety [of biconsonantal roots that lost their initial consonant] was largely increased in the descendant languages, notably Quenya, by the loss of older weak consonants initially: in Quenya, C.E. ʒ, h and g; in Telerin ʒ, ñ; in Sindarin h, were lost (VT48/26, Note 4).

This seems to imply that the voiced and voiceless [ɣ] and [x] were once again independent sounds rather than allophones in Primitive Elvish, since they had different developments in the child languages; it seems that in Quenya both sounds vanished initially at this stage. A similar set of distinct phonetic developments appears in another late (1969) note discussing the root √HAN, where Tolkien first wrote:

verbal stem ƷAN “give”, which in Q. and S. lost the initial spirant ʒ, that in T. became h- (PE22/163, note #99).

In a pencilled note, however, he added:

no. S. retained ʒ {but lost} and strengthened it to g. Q. lost ʒ - as also did T (PE22/163, note #99).

At this point, he revised the root from ƷAN >> HAN, and modified the phonetic description as follows:

verbal stem HAN “give”, which in S. lost the initial breath h of CE, that in Q./T. remained h- (PE22/163).

These notes seem to show initial ʒ- vanishing in Quenya, but h- surviving (and the reverse in Sindarin, with h- vanishing and ʒ- > g-). These examples from the late 1960s indicate that Tolkien considered ʒ and h to be distinct sounds in CE, with separate phonetic developments in the child languages, though he seemed to be wrestling with exactly what those developments were. Some further details on its primitive use appear in some hand-written notes in green ball-point pen from the second version of the Outline of Phonology (OP2); these green-pen notes were written around 1970 (PE19/73, second half of note #22):

(iv) The open consonant ʒ was the weak semivocalic element corresponding to the vowel position a, as y, w correspond to the vowel positions i and u. Originally it may not have been strictly a spirant, and no more frictional than y, w, being represented by mere vocalic hiatus or the clear beginning [’].

In the margin above this Tolkien wrote:

As far as Eldarin languages are concerned, however, its existence is simply theoretic and deduced systematically. There is no trace of such a consonant in Eldarin, unless it is concealed in some apparently monoconsontal bases as TĀ “high” < TAƷ‽

However, all of the above was crossed out in red ball-point with a note saying:

Transfer to Gen. Structure. No [ʒ] existed in Eldarin.

This may mean Tolkien abandoned the idea of an independent voiced velar spirant in Primitive Elvish, most likely going back to the idea of having only a voiceless velar spirant denoted “h”. @@@ transfer the above discussion to primitive entry on [ɣ].

Neo-Quenya: For the purposes of Neo-Quenya, I think it is best to ignore Tolkien’s reintroduction of independent ʒ and h in the late 1960s, especially since it isn’t entirely clear how he thought these sounds should develop (especially in Quenya). I would stick to the phonetic rules of the mid-1930s through mid-1960s, with a single sound ʒ/h disappearing everywhere in Sindarin, but becoming h- initially in Quenya. The exact primitive form of this sound is only of academic interest, since the end results of its phonetic development in the child languages are the same.