Skip to content

Commit

Permalink
Update convert.py
Browse files Browse the repository at this point in the history 
label.py と otoini.py の仕様変更に合わせて修正した。
  • Loading branch information
@oatsu-gh
oatsu-gh committed Sep 12, 2020
1 parent d5af3f9 commit f630554
Showing 1 changed file with 74 additions and 86 deletions.
160 changes: 74 additions & 86 deletions utaupy/convert.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -96,7 +96,7 @@ def ust2otoini_mono(ust, name_wav, d_table, dt=100, debug=False):
notes = ust.values

# UstのNoteオブジェクトごとにOtoオブジェクトを生成
l = [] # simple_otoを入れるリスト
kana_otoini = _otoini.OtoIni() # simple_otoを入れるリスト
t = 0 # ノート開始時刻を記録
for note in notes[2:-1]:
length = note.length_ms
Expand All @@ -108,12 +108,12 @@ def ust2otoini_mono(ust, name_wav, d_table, dt=100, debug=False):
simple_oto.preutterance = 0
simple_oto.consonant = length
simple_oto.cutoff = -length # 負で左ブランク相対時刻, 正で絶対時刻
l.append(simple_oto)
kana_otoini.append(simple_oto)
t += length # 今のノート終了位置が次のノート開始位置

# Otoを音素ごとに分割
new = [] # mono_otoを入れるリスト
for simple_oto in l:
mono_otoini = _otoini.OtoIni() # mono_otoを入れるリスト
for simple_oto in kana_otoini:
if debug:
print(f' {simple_oto.values}')
try:
Expand All @@ -128,75 +128,73 @@ def ust2otoini_mono(ust, name_wav, d_table, dt=100, debug=False):
# 子音+母音 「か(k a)」
if len(phonemes) == 2:
# 子音部分
first_oto = _otoini.Oto()
first_oto.filename = name_wav
first_oto.alias = phonemes[0]
first_oto.offset = simple_oto.offset - (2 * dt)
first_oto.overlap = 0
first_oto.preutterance = dt
first_oto.consonant = 2 * dt
first_oto.cutoff = - 2 * dt
new.append(first_oto)
oto = _otoini.Oto()
oto.filename = name_wav
oto.alias = phonemes[0]
oto.offset = simple_oto.offset - (2 * dt)
oto.overlap = 0
oto.preutterance = dt
oto.consonant = 2 * dt
oto.cutoff = - 2 * dt
mono_otoini.append(oto)
# 母音部分
second_oto = _otoini.Oto()
second_oto.filename = name_wav
second_oto.alias = phonemes[1]
second_oto.offset = simple_oto.offset - dt
second_oto.overlap = 0
second_oto.preutterance = dt
second_oto.consonant = 2 * dt
second_oto.cutoff = simple_oto.cutoff - dt
new.append(second_oto)
oto = _otoini.Oto()
oto.filename = name_wav
oto.alias = phonemes[1]
oto.offset = simple_oto.offset - dt
oto.overlap = 0
oto.preutterance = dt
oto.consonant = 2 * dt
oto.cutoff = simple_oto.cutoff - dt
mono_otoini.append(oto)
# 母音など 「あ(a)」「ん(cl)」「っ(cl)」「(pau)」「(br)」「(sil)」
elif len(phonemes) == 1:
first_oto = _otoini.Oto()
first_oto.filename = name_wav
first_oto.alias = phonemes[0]
first_oto.offset = simple_oto.offset - dt
first_oto.overlap = 0
first_oto.preutterance = dt
first_oto.consonant = 2 * dt
first_oto.cutoff = simple_oto.cutoff - dt
new.append(first_oto)
oto = _otoini.Oto()
oto.filename = name_wav
oto.alias = phonemes[0]
oto.offset = simple_oto.offset - dt
oto.overlap = 0
oto.preutterance = dt
oto.consonant = 2 * dt
oto.cutoff = simple_oto.cutoff - dt
mono_otoini.append(oto)
# 子音+半母音+母音 「ぐぁ(g w a)」
elif len(phonemes) == 3:
# 子音部分
first_oto = _otoini.Oto()
first_oto.filename = name_wav
first_oto.alias = phonemes[0]
first_oto.offset = simple_oto.offset - (2 * dt)
first_oto.overlap = 0
first_oto.preutterance = dt
first_oto.consonant = 2 * dt
first_oto.cutoff = - 2 * dt
new.append(first_oto)
oto = _otoini.Oto()
oto.filename = name_wav
oto.alias = phonemes[0]
oto.offset = simple_oto.offset - (2 * dt)
oto.overlap = 0
oto.preutterance = dt
oto.consonant = 2 * dt
oto.cutoff = - 2 * dt
mono_otoini.append(oto)
# 半母音部分
second_oto = _otoini.Oto()
second_oto.filename = name_wav
second_oto.alias = phonemes[1]
second_oto.offset = simple_oto.offset - dt
second_oto.overlap = 0
second_oto.preutterance = dt
second_oto.consonant = 2 * dt
second_oto.cutoff = - 2 * dt
new.append(second_oto)
oto = _otoini.Oto()
oto.filename = name_wav
oto.alias = phonemes[1]
oto.offset = simple_oto.offset - dt
oto.overlap = 0
oto.preutterance = dt
oto.consonant = 2 * dt
oto.cutoff = - 2 * dt
mono_otoini.append(oto)
# 母音部分
third_oto = _otoini.Oto()
third_oto.filename = name_wav
third_oto.alias = phonemes[2]
third_oto.offset = simple_oto.offset
third_oto.overlap = 0
third_oto.preutterance = dt
third_oto.consonant = 2 * dt
third_oto.cutoff = simple_oto.cutoff - (2 * dt)
new.append(third_oto)
oto = _otoini.Oto()
oto.filename = name_wav
oto.alias = phonemes[2]
oto.offset = simple_oto.offset
oto.overlap = 0
oto.preutterance = dt
oto.consonant = 2 * dt
oto.cutoff = simple_oto.cutoff - (2 * dt)
mono_otoini.append(oto)
else:
raise ValueError('len(alias) must be in [1, 2, 3]')

new[0].offset = 0
new[0].preutterance = 0
mono_otoini = _otoini.OtoIni()
mono_otoini.values = new
mono_otoini[0].offset = 0
mono_otoini[0].preutterance = 0
return mono_otoini


Expand All @@ -211,7 +209,7 @@ def ust2otoini_romaji_cv(ust, name_wav, d_table, dt=100, replace=True, debug=Fal
"""
ust.make_finalnote_R() # 最終ノートが休符じゃない場合を対策
notes = ust.values
l = [] # otoini生成元にするリスト
otoini = _otoini.OtoIni()
t = 0 # ノート開始時刻を記録

# NOTE: ここnotes[2:-1]とust.values[2:-1]で処理時間に差は出る？
Expand All @@ -221,7 +219,7 @@ def ust2otoini_romaji_cv(ust, name_wav, d_table, dt=100, replace=True, debug=Fal
try:
phonemes = d_table[note.lyric]
except KeyError as e:
print(f' [ERROR] KeyError in utaupy.convert.ust2otoini_romaji_cv : {e}')
print(f' [WARN] KeyError in utaupy.convert.ust2otoini_romaji_cv : {e}')
phonemes = note.lyric.split()

length = note.length_ms
Expand All @@ -246,21 +244,19 @@ def ust2otoini_romaji_cv(ust, name_wav, d_table, dt=100, replace=True, debug=Fal

# 4音素以上には未対応。特殊音素と判断して1音素として処理
else:
print('\nERROR when setting alias : phonemes = {}-------------'.format(phonemes))
print('\n[WARN] when setting alias : phonemes = {}-------------'.format(phonemes))
print('1エイリアスあたり 1, 2, 3 音素しか対応していません。')
oto.alias = ''.join(phonemes)
oto.overlap = 0

l.append(oto)
otoini.append(oto)
t += length # 今のノート終了位置が次のノート開始位置

# 最初が休符なことを想定して、
l[0].offset = 0 # 最初の左ブランクを0にする
l[0].overlap = 0 # 最初のオーバーラップを0にする
l[0].preutterance = 0 # 最初の先行発声を0にする
l[0].cutoff2 -= 2 * dt
otoini = _otoini.OtoIni()
otoini.values = l
otoini[0].offset = 0 # 最初の左ブランクを0にする
otoini[0].overlap = 0 # 最初のオーバーラップを0にする
otoini[0].preutterance = 0 # 最初の先行発声を0にする
otoini[0].cutoff2 -= 2 * dt
return otoini


Expand Down Expand Up @@ -303,25 +299,22 @@ def otoini2label(otoini, mode='auto', debug=False):
tmp.append([t_start, oto.alias])

# OtoオブジェクトをPhonemeオブジェクトに変換する
phonemes = []
label = _label.Label()
for i, v in enumerate(tmp[:-1]):
phoneme = _label.Phoneme()
phoneme.start = v[0]
phoneme.end = tmp[i + 1][0]
phoneme.symbol = v[1]
phonemes.append(phoneme)
label.append(phoneme)

# 最終Otoだけ終了位置が必要なので 特別な処理
oto = otoini_values[-1]
phoneme = _label.Phoneme()
phoneme.start = int(time_order_ratio * (oto.offset + oto.preutterance))
phoneme.end = int(time_order_ratio * oto.cutoff2) # 発声終了位置は右ブランク
phoneme.symbol = tmp[-1][1]
phonemes.append(phoneme)
label.append(phoneme)

# Labelクラスオブジェクト化
label = _label.Label()
label.values = phonemes
return label


Expand All @@ -335,9 +328,8 @@ def label2otoini(label, name_wav):
# time_order_ratio = label_time_order / otoini_time_order
time_order_ratio = 10**(-4)

l = [] # Otoオブジェクトを格納するリスト

# 各音素PhonemeオブジェクトをOtoオブジェクトに変換して、リストにまとめる。
otoini = _otoini.OtoIni()
# 各音素PhonemeオブジェクトをOtoオブジェクトに変換して、OtoIniに格納する。
for phoneme in label.values:
oto = _otoini.Oto()
oto.filename = name_wav
Expand All @@ -347,11 +339,7 @@ def label2otoini(label, name_wav):
oto.preutterance = 0.0
oto.consonant = (phoneme.end - phoneme.start) * time_order_ratio
oto.cutoff = - (phoneme.end - phoneme.start) * time_order_ratio
l.append(oto)

# Otoiniオブジェクト化
otoini = _otoini.OtoIni()
otoini.values = l
otoini.append(oto)
return otoini


Expand Down

0 comments on commit f630554

Please sign in to comment.