Do words differing only in the voicedness of consonants act as functionally homophonous?

Question

In Japanese, there are some sound differences that clearly distinguish words from one another, such as falling on a different 行 or 段 on the 五十音 ([傘]{かさ} vs. [風]{かぜ} or [友]{とも} vs その); while other sound differences make no difference as all (English 'l' vs. English 'r', vanishing <u> sounds, ヱビス vs. エビス).

Where on this spectrum does voicedness (e.g. さ vs. ざ) fall?

(I am aware that words are spelled one way or the other, but then you get situations like [出島]{でじま} vs. [広島]{ひろしま} where the 島 is clearly the same word, or the somewhat more extreme [仲田]{なかた} vs [仲田]{なかだ}.)

Answer 1

No, Japanese is full of voiced–unvoiced minimal pairs. There are minimal pairs even when the constituent morphemes are the same:

• 山川（やまかわ） mountains and rivers (dvandva compound)
• 山川（やまがわ） mountain rivers (modifier–head compound)

Dvandva compounds typically don't trigger rendaku, so the former is unvoiced. You can find plenty of other minimal pairs showing contrasts in voicedness, for example:

• 蚊（か） mosquito
• 蛾（が） moth

And so on. The important point is that native speakers hear these as clearly contrasting words, so it's a fundamental distinction in Japanese phonology. They are not homophonous in any sense.

Answer 2

This is coming from a purely linguistic point of view, but since there are minimal pairs between voiced and unvoiced sounds in Japanese, where words only differ by one consonant being voiced and one being unvoiced, MEANS that they are different phonemes. That is the linguistic definition of two sounds being different phonemes.

The fact that some kanji can be read with both a voiced and unvoiced variant due to rendaku does not change this, especially since the reading with rendaku and the reading without have different meanings, hence they are minimal pairs.

The fact that some words are homophones also has nothing to do with this.