Rendaku (連濁; Japanese pronunciation: [ɾendakɯ], lit. 'sequential voicing') is a pronunciation change seen in some compound words in Japanese. Rendaku modifies the consonant at the start of the second (or later) part of the compound, replacing a voiceless consonant, such as /k s t h/, with a voiced consonant, such as /ɡ z d b/. For example, the morpheme kami (paper) starts with the voiceless consonant /k/, which is replaced with the corresponding voiced consonant /ɡ/ in the compound word origami, from ori (fold) + kami.

Rendaku is common, but it does not occur in all compound words. A rule known as Lyman's law blocks rendaku when the second element already contains one of the voiced obstruent phonemes /ɡ z d b/, as in the compound word umikaze (sea breeze). Because the second element kaze (wind) contains /z/, its initial consonant /k/ remains voiceless. Rendaku is also blocked almost always when the second element of a compound is a recent loan into Japanese. Furthermore, rendaku may fail to occur even in contexts where no definite blocking factor is present.

In the Japanese writing system, rendaku affects how a morpheme is spelled when using one of the kana syllabaries: it causes the dakuten ("voicing mark", written as ゛) to be added to the upper right corner of the kana character that represents the first consonant and vowel in the second element of the compound. This is seen when comparing the hiragana spelling of kami (かみ) to that of origami (おりがみ): the kana character か (ka) in the first word is replaced with が (ga), with the dakuten, in the second. Rendaku is not marked in writing when a morpheme is spelled using kanji (logographs taken from Chinese characters). For example, kami (paper) is written with the kanji character 紙, which is unchanged when used in the spelling of origami (折り紙).

Linguistically, rendaku involves aspects of both pronunciation (phonology) and word structure (morphology); therefore, it is categorized as a morphophonological phenomenon.

Rendaku replaces a voiceless obstruent consonant with a voiced consonant sound. For example, the voiceless alveolar plosive [t] becomes the voiced alveolar plosive [d] in the context of rendaku. In this case, the new consonant retains the same manner and place of articulation as the original consonant. However, rendaku can also cause additional changes depending on the sounds involved.

In the context of Japanese phonology, some consonant sounds that seem distinct to English speakers, and that have distinct spellings in Hepburn romanization, are analyzed as allophones (contextual variants of a phoneme). The following table describes the effects of rendaku in the standard variety of Japanese, using both phonemic transcriptions (marked by slashes, such as /s/) and phonetic transcriptions (marked by square brackets, such as [ɕ]).

Depending on the accent and age of a speaker, the voiced velar nasal [ŋ] may be used in Japanese as an alternative to the voiced velar stop [ɡ], mainly in cases where the consonant occurs in the middle of a word. For speakers who use [ŋ] in the middle of words, [k] is replaced with [ŋ] in the context of rendaku. The sounds [ŋ] and [ɡ] are typically analyzed as allophones of a single phoneme /ɡ/ (although some phonologists have argued they are distinct phonemes for the minority of speakers who consistently distinguish them).

The voiceless affricates [ts tɕ] are commonly analyzed as allophones of /t/. Originally, [ts tɕ] corresponded to voiced affricate sounds [dz dʑ], whereas [s ɕ] corresponded to voiced fricative sounds [z ʑ]. However, the historical distinction between [dz dʑ] and [z ʑ] has been lost in the Tokyo-based standard (though not in all regional varieties of Japanese). The merged sounds may be pronounced either as voiced affricates or voiced fricatives, regardless of etymology (see yotsugana). Therefore, historical /du/ and /zu/ have merged as phonetic [(d)zɯ], and historical /di/ and /zi/ have merged as phonetic [(d)ʑi]. In the context of rendaku, dialects with the merger may be analyzed as possessing an underlying phonemic distinction between /du di/ and /zu zi/ that becomes neutralized on the phonetic level as [dzɯ dʑi]. Alternatively, Frellesvig 2010, p. 41 treats rendaku as replacing the phoneme /t/ with /z/ before /u i/. A third approach is taken by Vance 2022, who rejects the identification of [ts tɕ] as allophones of /t/, and instead postulates five voiceless phonemes /t ts s tɕ ɕ/ and three voiced phonemes /d (d)z (d)ʑ/, with rendaku turning /t/ into /d/, /ts s/ into /(d)z/, and /tɕ ɕ/ into /(d)ʑ/, respectively.