Establishing a phoneme inventory for Rau is difficult, given the complex phonetic alternations and the dialectal diversity. Rau has numerous dialects, each of which has a slightly different phonology. Rau dialects are generally mutually intelligible; the main differences are in the vowels and the voiced plosives. The most widely spoken dialects, including that of Bakhzon, have undergone many shifts in its consonants - most notably the total loss of non-nasal sonorants; additionally, Bakhzon has extensive vowel reduction.

Consonants

/p t k q/ <p t k q>
/b d̪ dz g ɢ/ <b d z g r>
/ɬ s ʃ x χ/ <l s sh kh h>
/m n ŋ/ <m n ng>
/ʇ ᵑʇ ᶢʇ/ <c nc gc>
/ǁ ᵑǁ ᶢǁ/ <x nx gx>
/ǂ ᵑǂ ᶢǂ/ <ç nç gç>
There are probably more clicks. (maybe Bakhzon merges a voiceless nasal series into plain)
/j w/ are marginally phonemic due to loanwords. Arguably, there is also /ɣ/, which, unlike the other velar consonants, is not prevelar.

/ɬ/ ranges from [ɬ~θ~t̪θ~t̪~f] in the various dialects; the value given here is the most common in Bâkhzon.
/x/ becomes [f] before rounded vowels in some dialects. /m/ becomes [f] after voiceless plosives in many dialects, including Bâkhzon.
The /d̪ t dz/ imbalance is addressed in various ways. Younger speakers may adopt the shift of ɬ > t̪ from the far-eastern dialects. Western dialects often have t dz > t̪ dʒ; far eastern dialects merge t dz to ɾ in most positions, but preserve dz as [z] where lenition is blocked.
/ɢ/ is almost always realized as [ʁ], even in dialects that have no other lenition of the voiced plosives.
Many dialects have /j w/.
Clicks were lost (or perhaps never existed) in inherited vocabulary, as is common in Hathic; there are a few cases where they're retained, but most of them are from loanwords. The coronal "harmonic clusters" of Amqolic are typically loaned as clicks; possibly there are also clicks with uvular release?

Vowels

Vowels vary extensively across dialects, even more so than consonants. The Bakhzon vowel inventory is:
/a e ʌ o i ɯ/ <a e u o i ou>

Pitch accent

In the Bakhzon dialect, there are three possible pitches on stress peaks: high, low, and rising. There are probably some details here that will be added later

Dialectal variance

- The merger of /ɒ/ and /a/, generally an eastern innovation
- The vaguely western chain shift of /ɒ o/ > /o u/
- The treatment of *[ð]: [ð], [d], [z], or [0]
- T-flapping, probably introduced through contact with Narng
- Syllabification of voiced plosives (Bakhzon is innovative here)
- Lenition of coda voiced plosives; certain dialects must now be analyzed as having contrastive vowel length
- /j w/, introduced in various places through loanwords, absent entirely in other places, and reinforced in others (there's probably a dialect where g > j)

Pan-Swampish Alphabet

Written somewhat phonetically.

/p t k q/ <p t k q>
/b d̪ dz g ɢ/ <b d z g r>
/ɬ s ʃ x χ/ <l s y x h>
/m n ŋ/ <m n ñ>
[ʲ] <i>
[æ a e ə ʌ o i ɯ] <c a e v v o i w>
Tones: high <u>, low <j>, rising <f>

Phonotactics

Pre-Rau allowed the plosive + liquid clusters /pl bl ml pr br mr kr gr ŋr/ in onset position. /pl bl ml/ became /pχ bɢ mɢ/; all other C2 liquids became /dz/. Outside clusters, the liquids /l r j/ became /ɢ dz g/.

(Note that lateral obstruents generally became velars, while the lateral liquid became a uvular. These are unrelated sound changes that happened at different times. Loss of lateral obstruents and θ (< *s) > ɬ are both recent and limited to the Bakhton area, as is ʁ (< *l) > ɢ.)

Harmonic cluster metathesis is more complicated. The languages of the Zhjumna Coast sprachbund (i.e. western Continental Hathic, most of Amqolic, and Narng, but generally not Kangshuic) all heavily restrict initial consonant clusters in ways unrelated to the sonority hierarchy. (In Amqoli, for example, onset stop + stop clusters must proceed from front to back, so /pt pk/ are allowed but */tp kp/ aren't.)

In Rau, cluster harmony applies to all clusters of two occlusives, and has the following rules:
1) A harmonic cluster may not contain two occlusives with the same articulator class. (i.e. labial, coronal, or dorsal)
2) A harmonic cluster may not contain a C1 labial.
3) A harmonic cluster may not contain a C2 velar.
4) The whole cluster takes on the MOA of C2.

In other words, the permissible cluster types are:
- TP TB TM
- KP KB KM
- KT KD KN
- the exceptional non-harmonic clusters /bɢ mɢ ndz Pdz Kdz/ (i.e. |bʁ mʁ nr Pr Kr|)

If a non-harmonic cluster would be created, it's resolved either by loss of the first consonant (if the two occlusives have the same articulator class), frication of the first consonant (if the two occlusives are both dorsal, in new close juncture), or metathesis.

When C2 is a labial, it's realized as [f]~[v]~[w]. The other class, dorsal + coronal, produces the clicks, which are dental, lateral, and palatal, and are uvularized if C1 is uvular. (Note that, because lateral obstruents have merged with velars and resolution rules of C1C1 clusters depend on juncture class, the sequence /k-k/ can give either [k], [kǂ], or [xk]. Neither laterals nor palatals are realized as such anymore - |ƛ λ c ɟ| are /k g s dz/.)

Additionally, morphology may create clusters of a stop and a resonant |l r j|, which are resolved as follows:
- If the resonant precedes the stop, the cluster undergoes metathesis.
- If the resonant is |j|, it fortites following a labial (i.e. |pj bj mj| /ps bdz mz/), drops following a coronal, and coalesces with a velar to form a palatal (i.e. |kj gj ŋj| > |c ɟ ɲɟ| > /s dz ndz/).
- If the resonant is |l|, it coalesces with a following coronal or dorsal, and otherwise (i.e. after a labial) behaves normally.
- If the resonant is |r|, it behaves normally (i.e. becomes /dz/).