In Primitive Elvish, voiced stops were unvoiced where they appeared before a voiceless stop, aspirate or [s]. Tolkien described this change in both versions of the Tengwesta Qenderinwa, from the 1930s (TQ1) and the 1950s (TQ2):

The voiced stops b, d, g, and the voiced spirant ʒ, were unvoiced before suffixal t, th, s (TQ1, PE18/54).

The voiced stops b, d, g and the (spirantal) continuant ʒ became unvoiced before suffixal th, t, s (TQ2, PE18/102).

This sound change was also mentioned in the Outline of Phonetic Development from the 1930s (OP1) and from the Outline of Phonology from the 1950s (OP2):

The voiced stops: b, d, g. On their prehistoric unvoicing, and consequent merging with p, t, k, before other voiceless consonants, especially t, s, see General Account (OP1, PE19/45).

The voiced stops: b, d, g. On their prehistoric unvoicing (and consequent merging with p, t, k) before voiceless consonants, chiefly the suffixal consonants t, th, s, see the General Account of Eldarin (OP2, PE19/90).

There is one statement in OP2 indicating this sound change took place in Common Eldarin (CE), and thus may not have applied to all Avarin languages:

Before s. In this position the primitive CE development, while b, d, g were still stops, was the unvoicing to p, t, k (OP2, PE19/95).

In Sindarin this phonetic development was obscured by the vocalization of spirants. It is more obvious in Quenya:

• ᴹ✶ab(a)sene- > Q. apsene- “remit, release, forgive” (VT43/18).
• ᴹ✶syadsē > syatsē > ᴹQ. hyatse “cleft, gash” (Ety/SYAD).
• ✶magta > makta > Q. mahta “to handle, wield” (VT47/18).
• ✶weg-tē [> *wektē] > Q. vehte “life” (PE17/189-190).

Because this phonetic development could obscure relationships between verb forms after voiced stops became spirants in Quenya, in some cases these forms were altered by grammatical analogy. This was discussed in OP2 in the 1950s:

The ancient forms were in current grammatical derivatives, made by devices still actively used or well-recognized, often set aside by analogy, since they frequently obscured the etymological relationships. New formations were produced, many even in early PQ, by the introduction of the medial forms of the consonants, as produced in normal intervocalic position, to stand before the suffix.

Thus: √ABA/BA: ab-ta > apta > aꝑþa > apta. This occurs, but became dissociated from the original past tense: avante “refused, denied, said nay”. For aꝑþa a new past tense aptane was made in TQ, and for avante a new present stem auta, in which b > ƀ > u̯ shows the same development as b before voiced medial consonants (such as r, l) (OP2, PE19/90).

A very similar quote appears in OP1 (PE19/45), with the added twist that in the 1930s, [pt] became [ps], so that verb was ᴹQ. apsa-.

Note that there is some evidence this sound change also applied to voice spirants, of which there was only one in CE: ʒ [ɣ]. Consider √WE’E > we’kā [weɣka?] > wekka > Q. vecca (PE17/190). Such combinations would be very rare, and Tolkien vacillated on whether ʒ [ɣ] was a phoneme in CE.

Conceptual Development: There is at least one example that hints that similar phonetic developments were in play in Tolkien’s earliest conception of the Elvish languages:

• ᴱ✶ndag+ta > ᴱQ. nahta- “to slay cruelly” (PE14/66).

Given the lack of examples, it’s hard to figure out the details of these developments in Early Period Elvish.

GS/§4.7 ds, dt, gt > ts, tt, kt @@@