Added accentification
Browse files- accentor.py +49 -0
- app.py +18 -5
accentor.py
ADDED
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from ukrainian_word_stress import Stressifier, StressSymbol
|
2 |
+
from ukrainian_accentor_transformer import Accentor
|
3 |
+
|
4 |
+
|
5 |
+
|
6 |
+
accentors = ["model", "vocab", "none"]
|
7 |
+
|
8 |
+
|
9 |
+
|
10 |
+
def stress_replace_and_shift(stressed: str):
|
11 |
+
stressed = stressed.replace(
|
12 |
+
StressSymbol.CombiningAcuteAccent, "+"
|
13 |
+
)
|
14 |
+
new_stressed = ""
|
15 |
+
start = 0
|
16 |
+
last = 0
|
17 |
+
|
18 |
+
# shift stress symbol by one "при+віт" -> "пр+ивіт"
|
19 |
+
while True:
|
20 |
+
plus_position = stressed.find("+", start)
|
21 |
+
if plus_position != -1:
|
22 |
+
new_stressed += (
|
23 |
+
stressed[last : plus_position - 1] + "+" + stressed[plus_position - 1]
|
24 |
+
)
|
25 |
+
start = plus_position + 1
|
26 |
+
last = start
|
27 |
+
else:
|
28 |
+
new_stressed += stressed[last:]
|
29 |
+
break
|
30 |
+
return new_stressed
|
31 |
+
|
32 |
+
|
33 |
+
|
34 |
+
stressify = Stressifier(stress_symbol=StressSymbol.CombiningAcuteAccent)
|
35 |
+
accentor_transformer = Accentor()
|
36 |
+
|
37 |
+
|
38 |
+
def accentification(sentence: str, mode: str):
|
39 |
+
sentence = sentence.replace("+", "")
|
40 |
+
sentence = sentence.replace(
|
41 |
+
StressSymbol.CombiningAcuteAccent, ""
|
42 |
+
)
|
43 |
+
if (mode == "vocab"):
|
44 |
+
accented_sentence = stressify(sentence)
|
45 |
+
elif (mode == "model"):
|
46 |
+
accented_sentence = accentor_transformer(sentence)
|
47 |
+
else:
|
48 |
+
accented_sentence = sentence
|
49 |
+
return accented_sentence
|
app.py
CHANGED
@@ -6,6 +6,7 @@ from torch import no_grad, package
|
|
6 |
import ctypes
|
7 |
import gc
|
8 |
|
|
|
9 |
|
10 |
|
11 |
config = {
|
@@ -38,14 +39,18 @@ def init_models():
|
|
38 |
return models
|
39 |
|
40 |
|
41 |
-
def tts(text: str, voice: str):
|
|
|
|
|
|
|
|
|
42 |
synt = models[voice]
|
43 |
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
|
44 |
with no_grad():
|
45 |
-
wav_data = synt.tts(
|
46 |
synt.save_wav(wav_data, fp)
|
47 |
trim_memory()
|
48 |
-
return fp.name
|
49 |
|
50 |
|
51 |
|
@@ -58,15 +63,23 @@ iface = gr.Interface(
|
|
58 |
inputs=[
|
59 |
gr.Textbox(
|
60 |
label="Input",
|
61 |
-
value="
|
62 |
),
|
63 |
gr.Radio(
|
64 |
label="Voice",
|
65 |
choices=voices,
|
66 |
value=voices[0],
|
67 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
],
|
69 |
-
outputs=gr.Audio(label="Output"),
|
70 |
title="🇺🇦 - Ukrainian Voices",
|
71 |
)
|
72 |
|
|
|
6 |
import ctypes
|
7 |
import gc
|
8 |
|
9 |
+
from accentor import accentification, stress_replace_and_shift, accentors
|
10 |
|
11 |
|
12 |
config = {
|
|
|
39 |
return models
|
40 |
|
41 |
|
42 |
+
def tts(text: str, voice: str, mode: str):
|
43 |
+
# accentor
|
44 |
+
accented_text = accentification(text, mode)
|
45 |
+
plussed_text = stress_replace_and_shift(accented_text)
|
46 |
+
# TTS
|
47 |
synt = models[voice]
|
48 |
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
|
49 |
with no_grad():
|
50 |
+
wav_data = synt.tts(plussed_text, **tts_kwargs)
|
51 |
synt.save_wav(wav_data, fp)
|
52 |
trim_memory()
|
53 |
+
return fp.name, accented_text
|
54 |
|
55 |
|
56 |
|
|
|
63 |
inputs=[
|
64 |
gr.Textbox(
|
65 |
label="Input",
|
66 |
+
value="Кам'янець-Подільський - місто в Хмельницькій області України, центр Кам'янець-Подільської міської об'єднаної територіальної громади і Кам'янець-Подільського району.",
|
67 |
),
|
68 |
gr.Radio(
|
69 |
label="Voice",
|
70 |
choices=voices,
|
71 |
value=voices[0],
|
72 |
),
|
73 |
+
gr.Radio(
|
74 |
+
label="Accentor",
|
75 |
+
choices=accentors,
|
76 |
+
value=accentors[0],
|
77 |
+
),
|
78 |
+
],
|
79 |
+
outputs=[
|
80 |
+
gr.Audio(label="Output"),
|
81 |
+
gr.Textbox(label="Stressed")
|
82 |
],
|
|
|
83 |
title="🇺🇦 - Ukrainian Voices",
|
84 |
)
|
85 |
|