Zamenhof described the Esperanto orthography as "one letter, one sound", but in terms of phonemics it's uncertain what this means: Are v and ŭ allophones? is voicing assimilation allowed, expected, or forbidden in sequences like kz? Zamenhof also did not specify Esperanto phonotactics, only saying that borrowings "need to conform to Esperanto orthography", and therefore spellings have been adopted that appear to violate his intentions, such as poŭpo, ŭato, jida, matĉo. However, many of these have proven to be untable, and have either fallen out of use or been replaced with pronunciations more in line with core Esperanto vocabulary.

The phonemic inventory

The Esperanto phonemes are rather similar to those of Polish, but are especially close to Belarusian, which as the language of the relatively tolerant, democratic, and pacifist Polish-Lithuanian Commonwealth, was historically and ideologically important to the creator of Esperanto.

The main differences among the consonants (compared to Belarusian) are the absense of palatalization in Esperanto, although this arguably survives marginally in the affectionate suffixes -nj- and -ĉj-, and the interjection tju!; and in the lack of a phonemic affricate /dz͡/, although again there are remnants in words such as edzo (spouse). [In Belarusian, the letters ł, l represent [l, l^{j], and so are covered by the absence of palatalization.] It can also be argued that the Esperanto sequences kv, gv are phonemic, representing the phonemic /kw, gw/ of the Latinate and Germanic source languages.}

On the other hand, the distinctions between several Esperanto consonants carry very light functional loads , though they are not in complementary distribution and therefore not allophones. The practical effect of this is that people who don't control these distinctions are still able to communicate without difficulty. These minor distinctions are ĵ [ʒ] vs. ĝ [dʒ͡], found in the suffix -aĵo (-ation) vs. aĝo (age); k [k] vs. ĥ [x] vs. h [h], found in koro (heart) vs. ĥoro (chorus) vs. horo (hour) and in the prefix ek- (inchoative) vs. eĥo (echo); and c [ts͡] vs. ĉ [tʃ͡], found in a few minimal pairs such as caro (tzar), ĉar (because); ci (thou), ĉi (proximate particle used with deictics); celo (goal), ĉelo (cell); -eco (-ness), eĉ (even); etc. For similar reasons, there is no distinction between [v] and [w], with v representing either.

Among the vowels, Esperanto lacks the Belarusian y [ɨ], which is relatively rare among the world's languages, but Belarusian does seem to have provided the diphthongs. Belarusian also seems to have been the model of the complementary distribution of v (restricted to the onset of a syllable), and ŭ (occuring only as a vocalic offglide), though this was modified slightly, with Belarusian oŭ corresponding to Esperanto ov (as in bovlo), and ŭ being restricted to the sequences aŭ, eŭ in Esperanto. [However, Zamenhof did allow initial ŭ in onomatopoeic words such as ŭa (wah!).] The semivowel j likewise does not occur before or after the vowel i. Later exceptions to these patterns, such as poŭpo (poop deck), ŭato (Watt), East Asian proper names beginning with Ŭ, and jida (Yiddish), are marginal.

The distinction between e and ej carries a light functional load, in the core vocabulary perhaps only distinctive before alveolar sonorants, such as kejlo (peg, spigot), kelo (cellar); mejlo (mile), melo (badger); Rejno (Rhine), reno (kidney). The recent borrowing gejo (gay, homosexual) would contrast with the ambisexual prefix ge- when used in compounds with a following consonant.

Stress is on the penultimate syllable, which each vowel defining a syllabic nucleus: familio [fa.mi.'li.o]] (family). The only exception is when the final -o of a noun is elided, usually for poetic reasons, as this does not affect the stress: famili' [fa.mi.'li].

Lexical tone is not phonemic, nor is clausal intonation, as question particles and changes in word order serve the functions that intonation perfoms in English.

Thus the consonantal inventory is as follows, with v counted as both a fricative and an approximant, parentheses around distinctions with low functional loads, and double parentheses around possible phonemes not represented by the orthography:

The vowels and diphthongs are:

Phonotactics

A syllable in Esperanto consists of an onset of 0-3 consonants, a nucleus of a single vowel or diphthong, and a coda of 0-1 (rarely 2) consonants. Any consonant may occur initially, with the exception of j before i [though there are now words that violate this restriction, such as jida (Yiddish) vs. ida (of an offspring)]. Any consonant but h may close a syllable, though coda ĝ and ĵ are rare or nonexistant in monomorphemes. Within a morpheme, this allows for a maximum of four sequential consonants, as for example in instrui (to teach).

Geminate consonants only occur in polymorphemic words, such as mal-longa (short), dis-sendi (distribute), and proper names such as Ŝillero (Schiller), Gallo (Gaul), Buddo (Buddha, more commonly Budho), with the exception of a couple late borrowings such as matĉo (match [in sports]).

Syllabic consonants occur only as interjections: fr!, sss!, ŝŝ!, hm!

Of the consonant clusters, several occur only in Greek roots or proper names, and this holds true for most two-consonant codas. Sfinktero (a sphincter) illustrates both. All triconsonantal onsets begin with a sibilant, s or ŝ. Disregarding proper names such as Vladimiro and Njemeno, the following initial consonant clusters occur:

Plosive + liquid — bl, br; pl, pr; dr; tr; gl, gr; kl, kr

Voiceless fricative + liquid — fl, fr; sl; ŝl, ŝr

Voiceless sibilant + voiceless plosive (+ liquuid) — sc [sts], sp, spl, spr; st, str; sk, skl, skr;    ŝp, ŝpr; ŝt, ŝtr

Obstruent + nasal — gn, kn, sm, ŝn, ŝm

Obstruent + /v/ — gv, kv, sv, ŝv

In addition, initial pf- occurs in German-derived pfenigo (penny), kŝ- in Sanscrit kŝatrio (kshatriya), and several additional uncommon initial clusters occur in technical words of Greek origin, such as mn-, pn-, ks-, ps-, ks-, sf-, ft-, kt-, pt-, bd-.

Many roots end in two or even three consonants, such as ŝultr-o (a shoulder), pingl-o (a needle), tranĉ-i (to cut). However, except for proper names and a few Greek roots, they do not produce coda clusters except when followed by another consonant in compounds, or with poetic elision of a final -o. Even then, only sequences with decreasing sonority are possible, so while poetic tranĉ' occurs, *ŝultr' and *pingl' do not.

Allophonic variation

With only five oral and no nasal or long vowels, Esperanto allows a fair amount of allophonic variation, though the distinction between /e/ and /ei̯/, and arguably /o/ and /ou̯/, is phonemic. Disregarding assimilation for the moment, the more noticeable allophony among the consonants is with /r/ and /v/. The /r/ may be pronounced as either an alveolar flap [ɾ] or an alveolar trill [r], in free variation but with the flap more common. The /v/ may be a labiodental fricative [v], a labiodental approximant [ʋ], or a labial-velar approximant [w], again in free variation, but with [v] considered normative, and [w] rare. Alveolar consonants t, d, n, l are acceptably either apical (as in English) or laminal (as in French, generally but incorrectly called "dental"). H may be voiced [ɦ], especially between vowels. However, aspiration or incomplete voicing of consonants in the English or Mandarin fashion is considered substandard, as are the English diphthongized "long" vowels [ij, ej, uw, ʌw].

Epenthetic glottal stops in vowel sequences such as boao (boa) are non-phonemic, but allowed for the comfort of the speaker. They are especially common with sequences of identical vowels, such as heroo (hero) and praa (archaic). Vowel elision may also occur, with a final o of a noun and the a of the article la being optional, though this rarely occurs outside of poetry.

An arguably allophonic aspect of Esperanto phonology is regressive voicing assimilation among obstruents, especially with the non-Latinate orthographic sequence kz found in frequent Latinate words like ekzemple (for example). It's generally considered proper to pronounce this with mixed voicing, [kz], as written, but many people argue the k should be assmilated to the z for [gz], as in Slavic, English, Spanish, French, and many other languages. The latter opinion is called egzismo in Esperanto. Tellingly, the question of assimilation is almost never raised with words such as absurda that maintain the Latinate orthography, despite the fact that potentially contrastive voiceless equivalents such as apsido (apsis) occur. [Orthographic gz does not occur in Esperanto, except in the nonce word egzismo itself.] However, voicing assimilation of fricatives, such as the s in the suffix -ismo, is both more noticeable and easier for most speakers to avoid, and so is less tolerated.

Like the generally ignored regressive devoicing in words like absurda, progressive devoicing tends to go unnoticed within plosive-sonorant clusters, as in plu (further; contrasts with blua blue) and knabo (boy; kn- contrasts with gn-, as in gnomo gnome). Partial or full devoicing of the sonorant is probably the norm for most speakers.

Similar arguments to the egzismo controversy arise in other cases where assimilation is familiar to European-language speakers, such as place assimilation of n before a velar (banko — [banko] or [baŋko]?). But again, the question almost never arises as to whether the m in emfazi should remain bilabial or should assimilate to labiodental f. Indeed, where the orthography allows, we see that assimilation does occur: for example, bonbono (bonbon) has become bombono even in dictionaries. In practice, most Esperanto speakers assimilate both kz to [gz] and nk to [ŋk] when speaking fluently.

Another source of variation is in proper names, when non-Esperantized remnants of the original orthography remain, or when odd sequences are created in order to avoid homonymy with existing roots. For example, it is doubtful that many people fully pronounce the g in Vaŝingtono (Washington), or the h in Budho (Buddha). Such situations are unstable, and in many cases dictionaries recognize that certain spellings (and therefore pronunciations) are inadvisable. For example, the physical unit "Watt" was first borrowed as ŭato, to distinguish it from vato (cotton-wool), and this is the only form found in dictionaries in 1930. However, this violates Esperanto phonotactics, and by 1970 there was an alternate spelling vatto. This was also unsatisfactory, however, due to the geminate t, and by 2000 the effort had been given up, with vato now the advised spelling for both "Watt" and "cotton-wool".

• See also Esperanto pronunciation.