[地下鉄]{ちかてつ}
How does one know the "hiragana spelling", i.e. the hiragana representation of the word is
ちかてつ
but not other forms involving misspelling one or more of the following
, especially when the pronunciation of か, て, つ in the word in reality is affected by a tendency to 濁音. I'm told that the unvoiced k-row sounds and t-row sounds are often pronounced as voiced g- and d- sounds when they are not appearing as the first sound of words.
Please enlighten me!
I think you've mixed up two different things:
The first one is a grammar rather than pronunciation. In most cases, whether the voicing happens or not is determined by word, and the kana spelling always reflects the pronunciation change i.e. 雲 (くも) → 雨雲 (あまぐも), 雷雲 (かみなりぐも), 浮き雲 (うきぐも) etc. You're not allowed to do such things like sometimes say あまぐも and sometimes あまくも.
Note that this rule refers historical classification of 清音/濁音, so we do 日 (ひ) → 夏日 (なつび), 曜日 (ようび) etc. even they no longer form voiced/unvoiced pair in today's pronunciation.
The second one is a pure pronunciation problem. We aspirate voiceless (stop) consonants in the beginning of word, but do not in other places: かた, とり (aspirated) vs. みかた, ことり (unaspirated). English speakers, too, aspirate them at the beginning of first or stressed syllables: pie, repine (aspirated) vs. spy, occupy (unaspirated). This doesn't mean it could change into voiced consonants. The voiceless/voiced contrast is fundamental in Japanese pronunciation.
However, if your mother tongue doesn't have this distinction, it's possible that you overlook the difference. Especially when you speak a language that counts aspiration but not voicedness (just opposite to Japanese), including Chinese, Korean, Georgian, Zulu or Scottish Gaelic, you may have false belief on sound qualities without proper care. If that's the case, what you have to do is just practice.
Technically, some languages like American English or German could have weaker voicing that the difference is mainly perceived through other factors. Japanese voicing I believe is more "typical" than them, but you can also use various side effects of voicing, such as buzzing undertone, slightly lower voice or less crisper sound, as aid of hearing.
Now I see that you have difficulties distinguishing voicedness.
Listen to it: https://clyp.it/0omton0o
I pronounced 12 pieces of sounds in total, with labial/dental/alveolar/velar consonants each combined with aspirated (VOT > 100ms), tenuis (0 < VOT < 10ms; don't mind tenseness) and voiced (VOT < -100ms) articulations.
aspirated tenuis voiced
p/b #1 #2 #3
t/d #4 #5 #6
ch/j #7 #8 #9
k/g #10 #11 #12
Basically, you have to train to accommodate yourself hearing #2, #5, #8, #11 as voiceless, and #3, #6, #9, #12 as voiced. However, there's another pitfall in Japanese, that people often use tenuis series instead of voiced series in word-initial, where unvoiced sounds are expected to be aspirated.
This post in linguistics SE would be helpful, too.
I don't agree with what you were told about it.
Of course in rapid speaking consonants tend to be pronounced rather roughly, and it may be prominent especially in non-initial consonants, but still voiceless consonants (k, t, ts, etc.) have to be distingushed from voiced ones (g, d, z, etc.) in Japanese.
For example in this conversation:
A: じゃあ明日には完成するんだな？
B: [誓]{ちか}うよ。
the word 誓う must not be pronounced ちがう, because ちがう means another word 違う, and, 誓う and 違う are totally different two words which are to be distinguished only by the voiced-ness of the second consonant, /k/ vs /g/.
In short (and linguistically), /k/ and /g/ are a pair of different phonemes which can constitute a "minimal pair".
Theoretically this kind of pair can happen with any word containing relevant consonants (like [香水]{こうすい} vs [洪水]{こうずい}, [囮]{おとり} vs [踊り]{おどり}, [文系]{ぶんけい} vs [文芸]{ぶんげい}..., etc.), so, I recommend you to always try to pronounce them distinctively; I mean, voiceless consonants voicelessly, and voiced voicedly, at least until you get very fluent in Japanese.
And all this situation also applies to any native speaker. They have to differentiate words in those pairs too. か/て/つ in ちかてつ is never pronounced fully voiced to appear as が/で/ず, so as to avoid confusing with words like ちがてつ, ちかでつ, ちがでず..., etc, though in this case they are all non-existent words.
[EDIT] In reply to qazwsx's comment
As for your final question, my answer is clear; Just forget this "g-ng contrasting" (which you are trying to perform) and simply stick to the "k-g contrasting" (read words as their ローマ字 are spelled). Have you ever seen a textbook that explains about this "g-ng contrasting" whatever? I bet you haven't (because there is no such one), and you should not take in such a rule which is not mentioned in any one of existing textbooks.
http://www.bunka.go.jp/kokugo_nihongo/joho/kakuki/20/tosin03/10.html
Here is a short report by 国語審議会 (National Language Council), stating that people incapable of pronouncing 鼻濁音 are increasing among young generation, in addition to the already existing people who use dialects without 鼻濁音. Your theory is WRONG in the first place.
To be very precise, yes, a kind of this "g-ng contrasting" does exist in Tohoku dialects, mainly among aged people. But when your are new to a language, no doubt you should learn the standardized one first, not a dialect.
So I strongly recommend you to quit this "g-ng" practice immediately, and get back to the standard "k-g" pronunciation.
(Expanding on my comments; this post only covers the spelling part rather than any pronunciation confusion you may have had.)
I'm told that the unvoiced k-row sounds and t-row sounds are often pronounced as voiced g- and d- sounds when they are not appearing as the first sound of words.
Kanji have on readings (音読み, Chinese-derived ones) and kun readings (訓読み, native Japanese ones). What you were told applies primarily only to kun readings. Rendaku happening with on readings is very rare. [1]
I admit that I don't know how the Chinese study (Japanese) kanji and their readings, but westerners typically learn the kanji and its readings, and they learn which readings are on readings, and which kun.
Also, as you may have learnt, words that typically consist of two kanji in a row are much more likely to use the on readings of a kanji.
Let's take 地下鉄, as it was your example. It's
地: on: チ ジ
下: on: カ ゲ kun: した しも もと さ.げる さ.がる くだ.る くだ.り くだ.す くだ.さる お.ろす お.りる
鉄: on: テツ kun: くろがね
You might note that チ、カ、 and テツ are all on readings; hence, there won't be any voicing for them. The word will be read as ちかてつ
Another word, 手紙:
手: on: シュ ズ kun: て
紙: on: シ kun: かみ
Note that かみ is a kun reading; hence, it is likely to rendaku into がみ in this word: てがみ
[生け]{いけ}[花]{ばな}, same thing. はな is a kun reading → likely rendaku.
[会]{エ}[釈]{シャク}:
会: カイ エ kun: あ.う あ.わせる あつ.まる
釈: シャク セキ kun: とく す.てる ゆる.す
シャク is an 音読み, so it won't be ジャク
Compound words, however, are a common exception to the 'no rendaku for on readings' rule. [会]{カイ}[社]{シャ} becomes がいしゃ when paired with the word 株式: [株]{かぶ}[式]{シキ} + [会]{ガイ}[社]{シャ}. My guess is that the Japanese no longer consider these at a kanji (reading) level but at a word level, and the word acts like a native kun reading.
[1] I'm seeing perhaps one case of a 連濁-ed on reading for every hundred words, if not fewer than that. [忍]{ニン}[者]{ジャ} is a good example
