Update app.py
Browse files
app.py
CHANGED
@@ -5,7 +5,6 @@ import re
|
|
5 |
import time
|
6 |
import uuid
|
7 |
from io import StringIO
|
8 |
-
|
9 |
import gradio as gr
|
10 |
import spaces
|
11 |
import torch
|
@@ -14,10 +13,10 @@ from huggingface_hub import HfApi, hf_hub_download, snapshot_download
|
|
14 |
from TTS.tts.configs.xtts_config import XttsConfig
|
15 |
from TTS.tts.models.xtts import Xtts
|
16 |
from vinorm import TTSnorm
|
|
|
17 |
|
18 |
# download for mecab
|
19 |
os.system("python -m unidic download")
|
20 |
-
|
21 |
HF_TOKEN = os.environ.get("HF_TOKEN")
|
22 |
api = HfApi(token=HF_TOKEN)
|
23 |
|
@@ -26,9 +25,7 @@ print("Downloading if not downloaded viXTTS")
|
|
26 |
checkpoint_dir = "model/"
|
27 |
repo_id = "capleaf/viXTTS"
|
28 |
use_deepspeed = False
|
29 |
-
|
30 |
os.makedirs(checkpoint_dir, exist_ok=True)
|
31 |
-
|
32 |
required_files = ["model.pth", "config.json", "vocab.json", "speakers_xtts.pth"]
|
33 |
files_in_dir = os.listdir(checkpoint_dir)
|
34 |
if not all(file in files_in_dir for file in required_files):
|
@@ -42,7 +39,6 @@ if not all(file in files_in_dir for file in required_files):
|
|
42 |
filename="speakers_xtts.pth",
|
43 |
local_dir=checkpoint_dir,
|
44 |
)
|
45 |
-
|
46 |
xtts_config = os.path.join(checkpoint_dir, "config.json")
|
47 |
config = XttsConfig()
|
48 |
config.load_json(xtts_config)
|
@@ -52,12 +48,10 @@ MODEL.load_checkpoint(
|
|
52 |
)
|
53 |
if torch.cuda.is_available():
|
54 |
MODEL.cuda()
|
55 |
-
|
56 |
supported_languages = config.languages
|
57 |
if not "vi" in supported_languages:
|
58 |
supported_languages.append("vi")
|
59 |
|
60 |
-
|
61 |
def normalize_vietnamese_text(text):
|
62 |
text = (
|
63 |
TTSnorm(text, unknown=False, lower=False, rule=True)
|
@@ -70,59 +64,52 @@ def normalize_vietnamese_text(text):
|
|
70 |
.replace("'", "")
|
71 |
.replace("AI", "Ây Ai")
|
72 |
.replace("A.I", "Ây Ai")
|
73 |
-
.replace("%"
|
74 |
)
|
75 |
return text
|
76 |
|
77 |
-
|
78 |
def calculate_keep_len(text, lang):
|
79 |
"""Simple hack for short sentences"""
|
80 |
if lang in ["ja", "zh-cn"]:
|
81 |
return -1
|
82 |
-
|
83 |
word_count = len(text.split())
|
84 |
num_punct = text.count(".") + text.count("!") + text.count("?") + text.count(",")
|
85 |
-
|
86 |
if word_count < 5:
|
87 |
return 15000 * word_count + 2000 * num_punct
|
88 |
elif word_count < 10:
|
89 |
return 13000 * word_count + 2000 * num_punct
|
90 |
return -1
|
91 |
|
92 |
-
|
93 |
@spaces.GPU
|
94 |
def predict(
|
95 |
prompt,
|
96 |
language,
|
97 |
audio_file_pth,
|
98 |
normalize_text=True,
|
|
|
|
|
99 |
):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
100 |
if language not in supported_languages:
|
101 |
metrics_text = gr.Warning(
|
102 |
-
f"Language you put {language} in is not in
|
103 |
)
|
104 |
-
|
105 |
return (None, metrics_text)
|
106 |
|
107 |
speaker_wav = audio_file_pth
|
108 |
-
|
109 |
if len(prompt) < 2:
|
110 |
metrics_text = gr.Warning("Please give a longer prompt text")
|
111 |
return (None, metrics_text)
|
112 |
|
113 |
-
# if len(prompt) > 250:
|
114 |
-
# metrics_text = gr.Warning(
|
115 |
-
# str(len(prompt))
|
116 |
-
# + " characters.\n"
|
117 |
-
# + "Your prompt is too long, please keep it under 250 characters\n"
|
118 |
-
# + "Văn bản quá dài, vui lòng giữ dưới 250 ký tự."
|
119 |
-
# )
|
120 |
-
# return (None, metrics_text)
|
121 |
-
|
122 |
try:
|
123 |
metrics_text = ""
|
124 |
t_latent = time.time()
|
125 |
-
|
126 |
try:
|
127 |
(
|
128 |
gpt_cond_latent,
|
@@ -133,7 +120,6 @@ def predict(
|
|
133 |
gpt_cond_chunk_len=4,
|
134 |
max_ref_length=60,
|
135 |
)
|
136 |
-
|
137 |
except Exception as e:
|
138 |
print("Speaker encoding error", str(e))
|
139 |
metrics_text = gr.Warning(
|
@@ -142,10 +128,8 @@ def predict(
|
|
142 |
return (None, metrics_text)
|
143 |
|
144 |
prompt = re.sub("([^\x00-\x7F]|\w)(\.|\。|\?)", r"\1 \2\2", prompt)
|
145 |
-
|
146 |
if normalize_text and language == "vi":
|
147 |
prompt = normalize_vietnamese_text(prompt)
|
148 |
-
|
149 |
print("I: Generating new audio...")
|
150 |
t0 = time.time()
|
151 |
out = MODEL.inference(
|
@@ -169,9 +153,7 @@ def predict(
|
|
169 |
# Temporary hack for short sentences
|
170 |
keep_len = calculate_keep_len(prompt, language)
|
171 |
out["wav"] = out["wav"][:keep_len]
|
172 |
-
|
173 |
torchaudio.save("output.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000)
|
174 |
-
|
175 |
except RuntimeError as e:
|
176 |
if "device-side assert" in str(e):
|
177 |
# cannot do anything on cuda device side error, need to restart
|
@@ -181,7 +163,6 @@ def predict(
|
|
181 |
)
|
182 |
gr.Warning("Unhandled Exception encounter, please retry in a minute")
|
183 |
print("Cuda device-assert Runtime encountered need restart")
|
184 |
-
|
185 |
error_time = datetime.datetime.now().strftime("%d-%m-%Y-%H:%M:%S")
|
186 |
error_data = [
|
187 |
error_time,
|
@@ -195,7 +176,6 @@ def predict(
|
|
195 |
write_io = StringIO()
|
196 |
csv.writer(write_io).writerows([error_data])
|
197 |
csv_upload = write_io.getvalue().encode()
|
198 |
-
|
199 |
filename = error_time + "_" + str(uuid.uuid4()) + ".csv"
|
200 |
print("Writing error csv")
|
201 |
error_api = HfApi()
|
@@ -205,7 +185,6 @@ def predict(
|
|
205 |
repo_id="coqui/xtts-flagged-dataset",
|
206 |
repo_type="dataset",
|
207 |
)
|
208 |
-
|
209 |
# speaker_wav
|
210 |
print("Writing error reference audio")
|
211 |
speaker_filename = error_time + "_reference_" + str(uuid.uuid4()) + ".wav"
|
@@ -216,19 +195,17 @@ def predict(
|
|
216 |
repo_id="coqui/xtts-flagged-dataset",
|
217 |
repo_type="dataset",
|
218 |
)
|
219 |
-
|
220 |
# HF Space specific.. This error is unrecoverable need to restart space
|
221 |
space = api.get_space_runtime(repo_id=repo_id)
|
222 |
if space.stage != "BUILDING":
|
223 |
api.restart_space(repo_id=repo_id)
|
224 |
else:
|
225 |
print("TRIED TO RESTART but space is building")
|
226 |
-
|
227 |
else:
|
228 |
if "Failed to decode" in str(e):
|
229 |
print("Speaker encoding error", str(e))
|
230 |
metrics_text = gr.Warning(
|
231 |
-
|
232 |
)
|
233 |
else:
|
234 |
print("RuntimeError: non device-side assert error:", str(e))
|
@@ -238,7 +215,7 @@ def predict(
|
|
238 |
return (None, metrics_text)
|
239 |
return ("output.wav", metrics_text)
|
240 |
|
241 |
-
|
242 |
with gr.Blocks(analytics_enabled=False) as demo:
|
243 |
with gr.Row():
|
244 |
with gr.Column():
|
@@ -288,6 +265,16 @@ with gr.Blocks(analytics_enabled=False) as demo:
|
|
288 |
info="Normalize Vietnamese text",
|
289 |
value=True,
|
290 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
291 |
ref_gr = gr.Audio(
|
292 |
label="Reference Audio (Giọng mẫu)",
|
293 |
type="filepath",
|
@@ -311,6 +298,8 @@ with gr.Blocks(analytics_enabled=False) as demo:
|
|
311 |
language_gr,
|
312 |
ref_gr,
|
313 |
normalize_text,
|
|
|
|
|
314 |
],
|
315 |
outputs=[audio_gr, out_text_gr],
|
316 |
api_name="predict",
|
|
|
5 |
import time
|
6 |
import uuid
|
7 |
from io import StringIO
|
|
|
8 |
import gradio as gr
|
9 |
import spaces
|
10 |
import torch
|
|
|
13 |
from TTS.tts.configs.xtts_config import XttsConfig
|
14 |
from TTS.tts.models.xtts import Xtts
|
15 |
from vinorm import TTSnorm
|
16 |
+
from content_generation import create_content # Nhập hàm create_content từ file content_generation.py
|
17 |
|
18 |
# download for mecab
|
19 |
os.system("python -m unidic download")
|
|
|
20 |
HF_TOKEN = os.environ.get("HF_TOKEN")
|
21 |
api = HfApi(token=HF_TOKEN)
|
22 |
|
|
|
25 |
checkpoint_dir = "model/"
|
26 |
repo_id = "capleaf/viXTTS"
|
27 |
use_deepspeed = False
|
|
|
28 |
os.makedirs(checkpoint_dir, exist_ok=True)
|
|
|
29 |
required_files = ["model.pth", "config.json", "vocab.json", "speakers_xtts.pth"]
|
30 |
files_in_dir = os.listdir(checkpoint_dir)
|
31 |
if not all(file in files_in_dir for file in required_files):
|
|
|
39 |
filename="speakers_xtts.pth",
|
40 |
local_dir=checkpoint_dir,
|
41 |
)
|
|
|
42 |
xtts_config = os.path.join(checkpoint_dir, "config.json")
|
43 |
config = XttsConfig()
|
44 |
config.load_json(xtts_config)
|
|
|
48 |
)
|
49 |
if torch.cuda.is_available():
|
50 |
MODEL.cuda()
|
|
|
51 |
supported_languages = config.languages
|
52 |
if not "vi" in supported_languages:
|
53 |
supported_languages.append("vi")
|
54 |
|
|
|
55 |
def normalize_vietnamese_text(text):
|
56 |
text = (
|
57 |
TTSnorm(text, unknown=False, lower=False, rule=True)
|
|
|
64 |
.replace("'", "")
|
65 |
.replace("AI", "Ây Ai")
|
66 |
.replace("A.I", "Ây Ai")
|
67 |
+
.replace("%", "phần trăm")
|
68 |
)
|
69 |
return text
|
70 |
|
|
|
71 |
def calculate_keep_len(text, lang):
|
72 |
"""Simple hack for short sentences"""
|
73 |
if lang in ["ja", "zh-cn"]:
|
74 |
return -1
|
|
|
75 |
word_count = len(text.split())
|
76 |
num_punct = text.count(".") + text.count("!") + text.count("?") + text.count(",")
|
|
|
77 |
if word_count < 5:
|
78 |
return 15000 * word_count + 2000 * num_punct
|
79 |
elif word_count < 10:
|
80 |
return 13000 * word_count + 2000 * num_punct
|
81 |
return -1
|
82 |
|
|
|
83 |
@spaces.GPU
|
84 |
def predict(
|
85 |
prompt,
|
86 |
language,
|
87 |
audio_file_pth,
|
88 |
normalize_text=True,
|
89 |
+
use_llm=False, # Thêm tùy chọn sử dụng LLM
|
90 |
+
content_type="Theo yêu cầu", # Loại nội dung (ví dụ: "triết lý sống" hoặc "Theo yêu cầu")
|
91 |
):
|
92 |
+
if use_llm:
|
93 |
+
# Nếu sử dụng LLM, tạo nội dung văn bản từ đầu vào
|
94 |
+
print("I: Generating text with LLM...")
|
95 |
+
generated_text = create_content(prompt, content_type, language)
|
96 |
+
print(f"Generated text: {generated_text}")
|
97 |
+
prompt = generated_text # Gán văn bản được tạo bởi LLM vào biến prompt
|
98 |
+
|
99 |
if language not in supported_languages:
|
100 |
metrics_text = gr.Warning(
|
101 |
+
f"Language you put {language} in is not in our Supported Languages, please choose from dropdown"
|
102 |
)
|
|
|
103 |
return (None, metrics_text)
|
104 |
|
105 |
speaker_wav = audio_file_pth
|
|
|
106 |
if len(prompt) < 2:
|
107 |
metrics_text = gr.Warning("Please give a longer prompt text")
|
108 |
return (None, metrics_text)
|
109 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
110 |
try:
|
111 |
metrics_text = ""
|
112 |
t_latent = time.time()
|
|
|
113 |
try:
|
114 |
(
|
115 |
gpt_cond_latent,
|
|
|
120 |
gpt_cond_chunk_len=4,
|
121 |
max_ref_length=60,
|
122 |
)
|
|
|
123 |
except Exception as e:
|
124 |
print("Speaker encoding error", str(e))
|
125 |
metrics_text = gr.Warning(
|
|
|
128 |
return (None, metrics_text)
|
129 |
|
130 |
prompt = re.sub("([^\x00-\x7F]|\w)(\.|\。|\?)", r"\1 \2\2", prompt)
|
|
|
131 |
if normalize_text and language == "vi":
|
132 |
prompt = normalize_vietnamese_text(prompt)
|
|
|
133 |
print("I: Generating new audio...")
|
134 |
t0 = time.time()
|
135 |
out = MODEL.inference(
|
|
|
153 |
# Temporary hack for short sentences
|
154 |
keep_len = calculate_keep_len(prompt, language)
|
155 |
out["wav"] = out["wav"][:keep_len]
|
|
|
156 |
torchaudio.save("output.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000)
|
|
|
157 |
except RuntimeError as e:
|
158 |
if "device-side assert" in str(e):
|
159 |
# cannot do anything on cuda device side error, need to restart
|
|
|
163 |
)
|
164 |
gr.Warning("Unhandled Exception encounter, please retry in a minute")
|
165 |
print("Cuda device-assert Runtime encountered need restart")
|
|
|
166 |
error_time = datetime.datetime.now().strftime("%d-%m-%Y-%H:%M:%S")
|
167 |
error_data = [
|
168 |
error_time,
|
|
|
176 |
write_io = StringIO()
|
177 |
csv.writer(write_io).writerows([error_data])
|
178 |
csv_upload = write_io.getvalue().encode()
|
|
|
179 |
filename = error_time + "_" + str(uuid.uuid4()) + ".csv"
|
180 |
print("Writing error csv")
|
181 |
error_api = HfApi()
|
|
|
185 |
repo_id="coqui/xtts-flagged-dataset",
|
186 |
repo_type="dataset",
|
187 |
)
|
|
|
188 |
# speaker_wav
|
189 |
print("Writing error reference audio")
|
190 |
speaker_filename = error_time + "_reference_" + str(uuid.uuid4()) + ".wav"
|
|
|
195 |
repo_id="coqui/xtts-flagged-dataset",
|
196 |
repo_type="dataset",
|
197 |
)
|
|
|
198 |
# HF Space specific.. This error is unrecoverable need to restart space
|
199 |
space = api.get_space_runtime(repo_id=repo_id)
|
200 |
if space.stage != "BUILDING":
|
201 |
api.restart_space(repo_id=repo_id)
|
202 |
else:
|
203 |
print("TRIED TO RESTART but space is building")
|
|
|
204 |
else:
|
205 |
if "Failed to decode" in str(e):
|
206 |
print("Speaker encoding error", str(e))
|
207 |
metrics_text = gr.Warning(
|
208 |
+
"It appears something wrong with reference, did you unmute your microphone?"
|
209 |
)
|
210 |
else:
|
211 |
print("RuntimeError: non device-side assert error:", str(e))
|
|
|
215 |
return (None, metrics_text)
|
216 |
return ("output.wav", metrics_text)
|
217 |
|
218 |
+
# Cập nhật giao diện Gradio
|
219 |
with gr.Blocks(analytics_enabled=False) as demo:
|
220 |
with gr.Row():
|
221 |
with gr.Column():
|
|
|
265 |
info="Normalize Vietnamese text",
|
266 |
value=True,
|
267 |
)
|
268 |
+
use_llm_checkbox = gr.Checkbox(
|
269 |
+
label="Sử dụng LLM để tạo nội dung",
|
270 |
+
info="Use LLM to generate content",
|
271 |
+
value=False,
|
272 |
+
)
|
273 |
+
content_type_dropdown = gr.Dropdown(
|
274 |
+
label="Loại nội dung",
|
275 |
+
choices=["triết lý sống", "Theo yêu cầu"],
|
276 |
+
value="Theo yêu cầu",
|
277 |
+
)
|
278 |
ref_gr = gr.Audio(
|
279 |
label="Reference Audio (Giọng mẫu)",
|
280 |
type="filepath",
|
|
|
298 |
language_gr,
|
299 |
ref_gr,
|
300 |
normalize_text,
|
301 |
+
use_llm_checkbox, # Thêm checkbox để bật/tắt LLM
|
302 |
+
content_type_dropdown, # Thêm dropdown để chọn loại nội dung
|
303 |
],
|
304 |
outputs=[audio_gr, out_text_gr],
|
305 |
api_name="predict",
|