Consonants:

Syllables with finals -u or -i (in Japanese often pron. as a "whisper" vowel) are used also for consonants, e.g.

ge-ru-ku > gerku (dog) - whereas (a fancy word like) greku would be gu-re-ku
shi-ri-no > shrino >crino (green) - whereas (a fancy word like) cirno would be shi-ru-no

But, since there exists consonant "n", one has two ways to go:

gu-nu-ka > {gunka} (work) or just gu-n-ka

Special initials:

According to Japanese phonetics, there are some specialties:

t-row: ta - chi (not: ti) - tsu (not tu) - te - to
d-row: da - dji (not: di) - dzu (not: du) - de - do
h-row: ha - hi - fu (not: hu) - he - ho
s-row: sa - shi (not: si) - su - se - so

So we have to build:

chi-a > tsha > tca; e.g. lojban: {tcadu} (city); chi-u > tshu > tcu etc.
tsu-a > tsa; e.g. lojban: {tsapi} (seasoning); tsu-i > tsi etc.
dji-a > dja; e.g. lojban {djacu} (water); dji-u > dju; e.g. lojban {djuno} (to know); etc.
dzu-a > dza; dzu-e > dze; e.g. lojban; {dzena} (elder) etc.
fu-a > fa; e.g. lojban: {fanza} (to annoy); fu-e > fe etc.
shi-a > sha > ca; lojban: {funca} (luck, fortune); shi-e > she > ce etc.

But we still need some more:

zu-shi-a > ja; lojban: {janbe} (bell); zu-shi-u > ju etc.
ku-ha > kha > xa; lojban: {xance} (hand); ku-hi > khi > xi; ku-hu/fu > khu > xu etc.
ya > ia; lojban attitudinals: {.ia} etc. (Since the characters for i and yi look the same, Lojban {.ii} should be given as double-i/yi)
wa > va; lojban: {vanbi} (environment); we > ve etc. (Since the characters for u and wu look the same, it could be advisable to give Lojban {vu} by wa-u)

'Double vowels' and diphtongs:

'Double vowels' (i.e. semi-vowel plus vowel) and diphtongs should be given by writing each single character,

e.g. e-i for {.ei}, u-u for {.uu} (not: wu - see above),
a-u for {.au},
u-a for {.ua},
i-i for {.ii} (not: yi - see above)
- But see above ya, yu, ye, yo for {.ia, .iu, .ie, .io}! -
etc.

Writing {cmene}:

cmene (names) should be written with a Japanese period (=circle) following the final consonant (i.e. the syllable having consonant value) or the final (open) syllable, indicating that there has to be added a virtual consonant final:

e.g. fu-ro-i-du. > {froid.} > Freud
e.g. ro-ma. > {romas.} > Roma/Rome (not: ro-ma-su.)
note that (German) "Rom" had to be given as ro-mu. -> {rom.} (!)
e.g. a-u-lu-n > {aulun.} > Aolung or a-u-lu-nu. (with period!)

Writing 'R' and 'L':

Japanese language does not differentiate between these two consonants nor does Hiragana . Lojban, on the other hand, has both 'R' and 'L'. So it would be advisable to create an L-row (e.g. by adding nigori - i.e. " - to the R-row characters - see above).

Writing the "schwa":

The "schwa" (lb: -y) is indicated by the Japanese (top-down) comma (see above)
e.g. la-mu-ne,-ge-ru-ku > {lamnygerku} ("sheep-dog" ->shepherd)

Writing Lojban " ' ":

The H-sound unique to Lojban (and only allowed between vowels) could easily be transcribed by using characters of the H-row:

e.g. ma-ha > {ma'a} (we = me and you and others);
mi-ho > {mi'o} (we = you and me);

but: di-hu/dji-fu > {di'u}/{djifu} ??? This points to a problem still unsolved! Or leave it ambiguous as it is - and kind of a riddle for machine translation :-)

Any suggestions?