Spaces:
Running
Running
Add num2words
Browse files- app.py +26 -11
- snfl_imdann.py +61 -0
app.py
CHANGED
@@ -3,6 +3,9 @@ import tempfile
|
|
3 |
from TTS.api import TTS
|
4 |
from huggingface_hub import hf_hub_download
|
5 |
import torch
|
|
|
|
|
|
|
6 |
|
7 |
CUDA = torch.cuda.is_available()
|
8 |
|
@@ -32,8 +35,26 @@ my_examples = [
|
|
32 |
["ⵜⴻⵜⵜⵏ ⴰⴳ ⵡⵓⵛⵛⵏ, ⵜⵜⵔⵓⵏ ⵅ ⵓⵎⴽⵙⴰ.", "rif", "idj", False]
|
33 |
]
|
34 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
my_inputs = [
|
36 |
-
gr.Textbox(lines=5, label="Input Text", placeholder=
|
37 |
gr.Dropdown(label="Variant", choices=list(VARIANTS.items()), value="shi"),
|
38 |
gr.Dropdown(label="Speaker", choices=SPEAKERS, value="yan"),
|
39 |
gr.Checkbox(label="Split Sentences (each sentence will be generated separately)", value=False),
|
@@ -43,15 +64,6 @@ my_inputs = [
|
|
43 |
|
44 |
my_outputs = gr.Audio(type="filepath", label="Output Audio", autoplay=True)
|
45 |
|
46 |
-
best_model_path = hf_hub_download(repo_id=REPO_ID, filename="checkpoint_390000.pth")
|
47 |
-
config_path = hf_hub_download(repo_id=REPO_ID, filename="config.json")
|
48 |
-
|
49 |
-
api = TTS(model_path=best_model_path, config_path=config_path).to("cuda" if CUDA else "cpu")
|
50 |
-
|
51 |
-
# pre-download voice conversion models
|
52 |
-
for model in VOICE_CONVERSION_MODELS.values():
|
53 |
-
api.load_vc_model_by_name(model, gpu=CUDA)
|
54 |
-
|
55 |
def tts(text: str, variant: str = "shi", speaker: str = "yan", split_sentences: bool = False, speaker_wav: str = None, voice_cv_model: str = 'freevc24'):
|
56 |
# replace oov characters
|
57 |
text = text.replace("\n", ". ")
|
@@ -62,6 +74,9 @@ def tts(text: str, variant: str = "shi", speaker: str = "yan", split_sentences:
|
|
62 |
text = text.replace(";", ",")
|
63 |
text = text.replace("-", " ")
|
64 |
|
|
|
|
|
|
|
65 |
with tempfile.NamedTemporaryFile(suffix = ".wav", delete = False) as fp:
|
66 |
if speaker_wav:
|
67 |
api.load_vc_model_by_name(VOICE_CONVERSION_MODELS[voice_cv_model], gpu=CUDA)
|
@@ -69,7 +84,7 @@ def tts(text: str, variant: str = "shi", speaker: str = "yan", split_sentences:
|
|
69 |
else:
|
70 |
api.tts_to_file(text, file_path=fp.name, split_sentences=split_sentences, speaker=speaker, language=variant)
|
71 |
|
72 |
-
return fp.name
|
73 |
|
74 |
iface = gr.Interface(
|
75 |
fn=tts,
|
|
|
3 |
from TTS.api import TTS
|
4 |
from huggingface_hub import hf_hub_download
|
5 |
import torch
|
6 |
+
import json
|
7 |
+
from snfl_imdann import TifinaghNumberConverter
|
8 |
+
import re
|
9 |
|
10 |
CUDA = torch.cuda.is_available()
|
11 |
|
|
|
35 |
["ⵜⴻⵜⵜⵏ ⴰⴳ ⵡⵓⵛⵛⵏ, ⵜⵜⵔⵓⵏ ⵅ ⵓⵎⴽⵙⴰ.", "rif", "idj", False]
|
36 |
]
|
37 |
|
38 |
+
best_model_path = hf_hub_download(repo_id=REPO_ID, filename="checkpoint_390000.pth")
|
39 |
+
config_path = hf_hub_download(repo_id=REPO_ID, filename="config.json")
|
40 |
+
|
41 |
+
api = TTS(model_path=best_model_path, config_path=config_path).to("cuda" if CUDA else "cpu")
|
42 |
+
|
43 |
+
# pre-download voice conversion models
|
44 |
+
for model in VOICE_CONVERSION_MODELS.values():
|
45 |
+
api.load_vc_model_by_name(model, gpu=CUDA)
|
46 |
+
|
47 |
+
with open(config_path, "r") as f:
|
48 |
+
config = json.load(f)
|
49 |
+
available_chars = config["characters"]["characters"]
|
50 |
+
available_punct = config["characters"]["punctuations"]
|
51 |
+
|
52 |
+
available_chars = available_chars + "".join([str(i) for i in range(10)])
|
53 |
+
|
54 |
+
placeholder = f"The available characters are: {available_chars} and the available punctuation is: {available_punct}"
|
55 |
+
|
56 |
my_inputs = [
|
57 |
+
gr.Textbox(lines=5, label="Input Text", placeholder=placeholder),
|
58 |
gr.Dropdown(label="Variant", choices=list(VARIANTS.items()), value="shi"),
|
59 |
gr.Dropdown(label="Speaker", choices=SPEAKERS, value="yan"),
|
60 |
gr.Checkbox(label="Split Sentences (each sentence will be generated separately)", value=False),
|
|
|
64 |
|
65 |
my_outputs = gr.Audio(type="filepath", label="Output Audio", autoplay=True)
|
66 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
def tts(text: str, variant: str = "shi", speaker: str = "yan", split_sentences: bool = False, speaker_wav: str = None, voice_cv_model: str = 'freevc24'):
|
68 |
# replace oov characters
|
69 |
text = text.replace("\n", ". ")
|
|
|
74 |
text = text.replace(";", ",")
|
75 |
text = text.replace("-", " ")
|
76 |
|
77 |
+
# convert numbers to their spoken form
|
78 |
+
text = re.sub(r"\d+", lambda x: TifinaghNumberConverter.convert(int(x.group(0))), text)
|
79 |
+
|
80 |
with tempfile.NamedTemporaryFile(suffix = ".wav", delete = False) as fp:
|
81 |
if speaker_wav:
|
82 |
api.load_vc_model_by_name(VOICE_CONVERSION_MODELS[voice_cv_model], gpu=CUDA)
|
|
|
84 |
else:
|
85 |
api.tts_to_file(text, file_path=fp.name, split_sentences=split_sentences, speaker=speaker, language=variant)
|
86 |
|
87 |
+
return fp.name
|
88 |
|
89 |
iface = gr.Interface(
|
90 |
fn=tts,
|
snfl_imdann.py
ADDED
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
class TifinaghNumberConverter:
|
2 |
+
AND = " ⴷ "
|
3 |
+
UNITS = ["ⴰⵎⵢⴰ", "ⵢⴰⵏ", "ⵙⵉⵏ", "ⴽⵕⴰⴹ", "ⴽⴽⵓⵥ", "ⵙⵎⵎⵓⵙ", "ⵚⴹⵉⵚ", "ⵙⴰ", "ⵜⴰⵎ", "ⵜⵥⴰ"]
|
4 |
+
TENS = ["", "ⵎⵔⴰⵡ", "ⵙⵉⵎⵔⴰⵡ", "ⴽⵕⴰⵎⵔⴰⵡ", "ⴽⴽⵓⵎⵔⴰⵡ", "ⵙⵎⵎⵓⵎⵔⴰⵡ", "ⵚⴹⵉⵎⵔⴰⵡ", "ⵙⴰⵎⵔⴰⵡ", "ⵜⴰⵎⵔⴰⵡ", "ⵜⵥⴰⵎⵔⴰⵡ"]
|
5 |
+
HUNDREDS = ["", "ⵜⵉⵎⵉⴹⵉ", "ⵙⵏⴰⵜ ⵜⵎⴰⴹ", "ⴽⵕⴰⴹⵜ ⵜⵎⴰⴹ", "ⴽⴽⵓⵥⵜ ⵜⵎⴰⴹ", "ⵙⵎⵎⵓⵙⵜ ⵜⵎⴰⴹ", "ⵚⴹⵉⵚⵜ ⵜⵎⴰⴹ", "ⵙⴰⵜ ⵜⵎⴰⴹ", "ⵜⴰⵎⵜ ⵜⵎⴰⴹ", "ⵜⵥⴰⵜ ⵜⵎⴰⴹ"]
|
6 |
+
ORDERS = ["", "ⵉⴼⴹ", "ⴰⵎⵍⵢⵓⵏ", "ⴰⵎⵍⵢⴰⵕ", "ⴰⵜⵔⵉⵍⵢⵓⵏ", "ⴰⴽⵡⴰⴹⵕⵉⵍⵢⵓⵏ", "ⴰⴽⵡⵉⵏⵜⵔⵉⵍⵢⵓⵏ", "ⴰⵙⵉⴽⵙⵜⵉⵍⵢⵓⵏ", "ⴰⵙⵉⴱⵜⵉⵍⵢⵓⵏ", "ⴰⵡⴽⵜⵉⵍⵢⵓⵏ", "ⴰⵏⵓⵏⵉⵍⵢⵓⵏ"]
|
7 |
+
ORDERS_PLURAL = ["", "ⵡⴰⴼⴹⴰⵏ", "ⵉⴷ ⴰⵎⵍⵢⵓⵏ", "ⵉⴷ ⴰⵎⵍⵢⴰⵕ", "ⵉⵜⵔⵉⵍⵢⵓⵏⵏ", "ⵉⴽⵡⴰⴹⵕⵉⵍⵢⵓⵏⵏ", "ⵉⴽⵡⵉⵏⵜⵔⵉⵍⵢⵓⵏⵏ", "ⵉⵙⵉⴽⵙⵜⵉⵍⵢⵓⵏⵏ", "ⵉⵙⵉⴱⵜⵉⵍⵢⵓⵏⵏ", "ⵉⵡⴽⵜⵉⵍⵢⵓⵏⵏ", "ⵉⵏⵓⵏⵉⵍⵢⵓⵏⵏ"]
|
8 |
+
|
9 |
+
@classmethod
|
10 |
+
def convert_number_to_999(cls, n):
|
11 |
+
"""Convert numbers from 0 to 999 to Tifinagh."""
|
12 |
+
if n == 0:
|
13 |
+
return ""
|
14 |
+
if n < 10:
|
15 |
+
return cls.UNITS[n]
|
16 |
+
if n < 20:
|
17 |
+
if n % 10 == 0:
|
18 |
+
return cls.TENS[n // 10]
|
19 |
+
return cls.UNITS[n % 10] + cls.AND + cls.TENS[n // 10]
|
20 |
+
if n < 100:
|
21 |
+
if n % 10 == 0:
|
22 |
+
return cls.TENS[n // 10]
|
23 |
+
return cls.TENS[n // 10] + cls.AND + cls.UNITS[n % 10]
|
24 |
+
if n < 1000:
|
25 |
+
if n % 100 == 0:
|
26 |
+
return cls.HUNDREDS[n // 100]
|
27 |
+
return cls.HUNDREDS[n // 100] + cls.AND + cls.convert_number_to_999(n % 100)
|
28 |
+
return ""
|
29 |
+
|
30 |
+
@classmethod
|
31 |
+
def convert_large_number(cls, n):
|
32 |
+
"""Convert large numbers to Tifinagh."""
|
33 |
+
if n < 1000:
|
34 |
+
return cls.convert_number_to_999(n)
|
35 |
+
|
36 |
+
for order in range(len(cls.ORDERS) - 1, 0, -1):
|
37 |
+
order_value = 10 ** (order * 3)
|
38 |
+
if n >= order_value:
|
39 |
+
quotient = n // order_value
|
40 |
+
remainder = n % order_value
|
41 |
+
|
42 |
+
if quotient == 1:
|
43 |
+
if remainder:
|
44 |
+
return cls.ORDERS[order] + cls.AND + cls.convert_large_number(remainder)
|
45 |
+
return cls.ORDERS[order]
|
46 |
+
else:
|
47 |
+
if remainder:
|
48 |
+
return cls.convert_number_to_999(quotient) + " " + \
|
49 |
+
cls.ORDERS_PLURAL[order] + cls.AND + cls.convert_large_number(remainder)
|
50 |
+
return cls.convert_number_to_999(quotient) + " " + cls.ORDERS_PLURAL[order]
|
51 |
+
|
52 |
+
return ""
|
53 |
+
|
54 |
+
@classmethod
|
55 |
+
def convert(cls, number):
|
56 |
+
"""Main conversion method."""
|
57 |
+
if number == 0:
|
58 |
+
return "ⴰⵎⵢⴰ"
|
59 |
+
if number < 0:
|
60 |
+
return "- " + cls.convert_large_number(abs(number))
|
61 |
+
return cls.convert_large_number(number)
|