1. The word [yeZH3di] occurs 36 secs after the song begins.
2. The most succint ZH in the song occurs as part of the word
[toZHikk3] exactly 4:16 secs into the song.
Oh, and, Piotr, you were right about it merging in with the retroflex
L in modern Tamil. Though I don't know whether this phenomenon is
peculiar to any particular dialect of Tamil: one of my malayali
friends claims that in fast speech his retroflex L's lapse into ZH's
because the ZH's are so much *easier* to produce ("isn't it ironic"),
in which case it might well be a semivocalic L.
PS: Say, how did you like the song? You can find the lyrics translated
and crudely transliterated at