In Quenya any initial voiceless hm, nn (derived from sm, sn) were voiced to m, n, and the same was generally true of voiceless initial hr, hl (derived from sr, sl). This is the system as described in the Outline of Phonetic Development (OP1) from the 1930s:

In AQ and in early classical PQ these sounds still appeared represented by signs that may be transcribed hm, hn. But later in classical PQ initial hm, hn become normal m, n (OP1: PE19/37).

hr, hl. These remained the genuine classical PQ forms, and are retained still in correct and careful spelling. But since hr, hl > r, l already in later pronunciation even of the classical period in TQ hr, hl are usually represented by r, l and in spelling the distinction is often neglected. The Noldor possessed initial voiceless ľ, ř and hence often retained the early classical pronunciation. The distinction between voiceless l, r and voiced was always made by the lore-masters or in ceremonial recitations of ancient texts (OP1: PE19/37).

The voicing of voiceless nasals was in (early) Parmaquesta [PQ], but the voicing of liquids was not fully realized until Tarquesta [TQ], and only fully in the Lindarin (pre-Vanyarin) dialect. This system is well-reflected in The Etymologies from the 1930s, where we consistently see sm-, sn-, sl-, sr- (> hm-, hn-, hl-, hr-) > m-, n-, l-, r-:

• ᴹ✶smaldā [> *hmalda] > ᴹQ. malda “gold (as metal)” (Ety/SMAL).
• ᴹ√SNEW [> *hneuma] > ᴹQ. neuma “snare” (Ety/SNEW).
• ᴹ✶slaiwā [> *hlaiwa] > ᴹQ. laiwa “sickly, sick, ill” (Ety/SLIW).
• ᴹ√SRI [> *hríma] > ᴹQ. ríma “edge, hem, border”, a proposed alteration to the root ᴹ√RĪ (Ety/RĪ; EtyAC/RĪ).

Tolkien described essentially the same system in the Outline of Phonology (OP2) from the 1950s, except that the voicing of hr, hl was shifted to the Ñoldorin dialect, and presumably the unvoiced sounds were retained in Vanyarin:

This voiceless m̌, ň was still present in AQ & was represented by signs that may be transcribed hm, hn. These voiceless nasals, limited to initial position and not of very frequent occurrence, became normal m, n (OP2: P19/79).

Among the Noldor hr, hl became voiced to r, l before the Exile, and the use of r, l in these cases was normal in TQ, as spoken, tho’ the spelling was usually maintained. Since later the Exiles were familiar with voiceless hr, hl in their Sindarin speech many of them restored this sound in TQ, according to the traditional spelling. The learned had, of course, at all times retained hr, hl in reading or reciting PQ (OP2: P19/79).

In The Lord of the Rings Appendix E Tolkien said that hl- became l- in pronunciation, and hinted that the same was true of hr-:

LH represents this sound when voiceless (usually derived from initial sl-). In (archaic) Quenya this is written hl, but was in the Third Age usually pronounced as l (LotR/1114).

RH represents a voiceless r (usually derived from older initial sr-). It was written hr in Quenya. Cf. L (LotR/1114).

Although it isn’t clear, the “Cf. L” cross reference may indicate that hr, like hl, became voiced in Quenya as described in OP1 and OP2. Thus, it seems the basic system remained the same from the 1930s forward. Starting with the 1940s, however, we see initial hr-, hl- appearing in some words:

• ᴹQ. hríve “winter” (Lord of the Rings drafts, PM/134).
• ᴹQ. hlasta- “hear” (Quenya Verbal System, PE22/103).
• ᴹQ. hlike “creeps” (Quenya Verbal System, PE22/113).

In the 1950s and 60s, words with initial hr-, hl- were very common:

• √SRAG > Q. hranga “difficult” (PE17/154, 185).
• ✶srāwe > Q. hrávë “flesh” (MR/349-350).
• √SLAS > Q. hlas “ear” (PE17/62).
• √SLON > Q. hlōn “sound” (VT48/29).

The last example provides some insight into the medial development of sl, sr. A related quote from the late 1960s appearing in the same document is:

hl, hr, hw, hy. In compounds with clearly perceived prefixes (or between clearly analyzed and separate other elements) these remain (as voiceless l, r, w, y) as in ohlon (not ollon) “diphthong” from ŏ + hlōn “sound” (< *slōn) (VT48/29).

Thus it seems that the hl, hr could be preserved (or restored) in Tarquesta [TQ] compounds or grammatical formations, as opposed to the usual medial development of sl, sr > zl, zr > ll, rr. As pointed out by Patrick Wynne (VT48/34 note #29) there are examples of medial sr > rr in ancient compounds such as ✶mi-srawanwe > Q. Mirröanwi “Incarnates” (MR/350). Wynne likewise suggested that the compounds ᴹQ. Meterríve and Norríve containing ᴹQ. hríve (PM/135) may be a counterexample of the restoration of medial hr in TQ compounds, since as names of months they are unlikely to be ancient. However, there are other examples from the Quenya Verbal System from 1948 (published after Wynne did his analysis) that clearly show the restoration of medial hl in verbal perfects:

• ahlázie perfect of ᴹQ. hlasta- (PE22/103).
• ihlíkie perfect of ᴹQ. hlik-, replacing archaic †illīk[ie] (PE22/113).

These perserved medial hl only make sense if the initial hl was itself not lost or was restored, meaning Tolkien may have been uncertain of the exact developments for initial hl, hr.

Conceptual Development: As discussed in the entry on how initial [s] unvoiced following continuant, in Early Qenya initial sm-, sn-, sl-, sr- all resulted in voiced m-, n-, l-, r-. See that entry for further details.

Neo-Quenya: Given the prevalence of initial hl-, hr- forms in the 1950s and 60s, I would retain such words, and would also revise words from the 1930s derived from sl-, sr- to begin with the voiceless liquids, at least in spelling. Although I would use hl-, hr- in spelling, I would treat both the voiced and voiceless variants as acceptable pronunciations, given the rules laid out by Tolkien in The Lord of the Rings Appendix E (and frankly because voiceless liquids can be hard to pronounce).