From The Sounds of Japanese (Vance 2008), p.85:
We'll transcribe [dz] phonemically as /z/ because there's no contrast between [dz] and the voiced lamino-alveolar fricative [z]. Typically, though not consistently, [dz] occurs at the beginning of a word or in the middle of a word immediately following a syllable-final consonant (§5.1, §5.6), and [z] occurs in the middle of a word immediately following a vowel. In short, [dz] and [z] are allophones of this /z/ phoneme. Most native speakers of Japanese are quite surprised to discover there's actually a phonetic difference to worry about, but you'll hear it if you listen carefully to pronunciations of zu [dzɯ] 図 'diagram' and chizu [cɕizɯ] 地図 'map'.
In the above, "a syllable-final consonant" means either
ん /N/ or
っ /Q/, which Vance explains in sections §5.1 and §5.6 respectively. In the latter section (p.108), he goes on to write:
As we saw in §4.3, /z/ has both [z] and [dz] as careful-pronunciation allophones, but following /Q/, /z/ is always [dz].
Although note that voiced geminates like this appear almost exclusively in loanwords, and even there under certain conditions are commonly devoiced—see e.g. A corpus-based study of geminate devoicing in Japanese (Kawahara and Sano 2013) or their other recent work for discussion.