Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -7,10 +7,11 @@ import librosa
|
|
7 |
import soundfile as sf
|
8 |
from pydub import AudioSegment
|
9 |
from telegram import Update
|
10 |
-
from telegram.ext import ApplicationBuilder, MessageHandler, filters
|
11 |
from transformers import pipeline, AutoTokenizer, VitsModel
|
12 |
from huggingface_hub import login
|
13 |
import asyncio
|
|
|
14 |
|
15 |
# ===== تهيئة التوكن =====
|
16 |
login(token=os.getenv("HF_TOKEN"))
|
@@ -22,20 +23,22 @@ logging.basicConfig(
|
|
22 |
)
|
23 |
logger = logging.getLogger(__name__)
|
24 |
|
25 |
-
# ===== تحميل النماذج =====
|
26 |
try:
|
|
|
27 |
asr_pipeline = pipeline(
|
28 |
"automatic-speech-recognition",
|
29 |
model="jonatasgrosman/wav2vec2-large-xlsr-53-arabic",
|
30 |
token=os.getenv("HF_TOKEN")
|
31 |
)
|
32 |
|
|
|
33 |
tts_tokenizer = AutoTokenizer.from_pretrained(
|
34 |
-
"
|
35 |
token=os.getenv("HF_TOKEN")
|
36 |
)
|
37 |
tts_model = VitsModel.from_pretrained(
|
38 |
-
"
|
39 |
token=os.getenv("HF_TOKEN")
|
40 |
)
|
41 |
|
@@ -43,13 +46,16 @@ except Exception as e:
|
|
43 |
logger.error(f"فشل تحميل النماذج: {str(e)}")
|
44 |
raise
|
45 |
|
|
|
|
|
|
|
46 |
# ===== دوال معالجة الصوت =====
|
47 |
def enhance_audio(input_path: str, output_path: str) -> bool:
|
48 |
try:
|
49 |
audio = AudioSegment.from_wav(input_path)
|
50 |
-
audio = audio.low_pass_filter(3000)
|
51 |
-
audio = audio.high_pass_filter(100)
|
52 |
-
audio = audio.normalize()
|
53 |
audio = audio.fade_in(150).fade_out(150)
|
54 |
audio.export(output_path, format="wav")
|
55 |
return True
|
@@ -67,16 +73,22 @@ async def speech_to_text(audio_path: str) -> str:
|
|
67 |
logger.error(f"فشل التعرف على الصوت: {str(e)}")
|
68 |
return ""
|
69 |
|
70 |
-
async def generate_response(text: str) -> str:
|
|
|
71 |
try:
|
|
|
|
|
|
|
|
|
72 |
chatbot = pipeline(
|
73 |
"text-generation",
|
74 |
model="aubmindlab/aragpt2-base",
|
75 |
-
token=os.getenv("HF_TOKEN")
|
|
|
|
|
76 |
)
|
77 |
response = chatbot(
|
78 |
-
|
79 |
-
max_length=100,
|
80 |
num_return_sequences=1,
|
81 |
pad_token_id=50256
|
82 |
)
|
@@ -86,25 +98,33 @@ async def generate_response(text: str) -> str:
|
|
86 |
return "حدث خطأ في توليد الرد."
|
87 |
|
88 |
async def text_to_speech(text: str) -> None:
|
|
|
89 |
try:
|
90 |
inputs = tts_tokenizer(text, return_tensors="pt")
|
91 |
with torch.no_grad():
|
92 |
-
output = tts_model(**inputs)
|
93 |
waveform = output.waveform[0].numpy()
|
94 |
sf.write("bot_response.wav", waveform, tts_model.config.sampling_rate)
|
95 |
except Exception as e:
|
96 |
logger.error(f"فشل تحويل النص إلى صوت: {str(e)}")
|
97 |
|
98 |
-
# =====
|
|
|
|
|
|
|
|
|
99 |
async def process_voice(update: Update, context):
|
100 |
try:
|
|
|
101 |
voice_file = await update.message.voice.get_file()
|
102 |
await voice_file.download_to_drive("user_voice.ogg")
|
103 |
|
|
|
104 |
user_text = await speech_to_text("user_voice.ogg")
|
105 |
-
bot_response = await generate_response(user_text)
|
106 |
await text_to_speech(bot_response)
|
107 |
|
|
|
108 |
if enhance_audio("bot_response.wav", "bot_response_enhanced.wav"):
|
109 |
await update.message.reply_voice("bot_response_enhanced.wav")
|
110 |
else:
|
@@ -114,11 +134,13 @@ async def process_voice(update: Update, context):
|
|
114 |
logger.error(f"خطأ غير متوقع: {str(e)}")
|
115 |
await update.message.reply_text("⚠️ عذرًا، حدث خطأ في المعالجة.")
|
116 |
|
|
|
117 |
def run_bot():
|
118 |
loop = asyncio.new_event_loop()
|
119 |
asyncio.set_event_loop(loop)
|
120 |
|
121 |
application = ApplicationBuilder().token(os.getenv("TELEGRAM_TOKEN")).build()
|
|
|
122 |
application.add_handler(MessageHandler(filters.VOICE, process_voice))
|
123 |
|
124 |
application.run_polling(
|
@@ -127,7 +149,6 @@ def run_bot():
|
|
127 |
)
|
128 |
|
129 |
if __name__ == "__main__":
|
130 |
-
# تشغيل البوت في خيط منفصل
|
131 |
bot_thread = threading.Thread(target=run_bot, daemon=True)
|
132 |
bot_thread.start()
|
133 |
bot_thread.join()
|
|
|
7 |
import soundfile as sf
|
8 |
from pydub import AudioSegment
|
9 |
from telegram import Update
|
10 |
+
from telegram.ext import ApplicationBuilder, MessageHandler, filters, CommandHandler
|
11 |
from transformers import pipeline, AutoTokenizer, VitsModel
|
12 |
from huggingface_hub import login
|
13 |
import asyncio
|
14 |
+
from collections import defaultdict
|
15 |
|
16 |
# ===== تهيئة التوكن =====
|
17 |
login(token=os.getenv("HF_TOKEN"))
|
|
|
23 |
)
|
24 |
logger = logging.getLogger(__name__)
|
25 |
|
26 |
+
# ===== تحميل النماذج مع صوت أنثوي محسّن =====
|
27 |
try:
|
28 |
+
# 1. نموذج التعرف على الكلام
|
29 |
asr_pipeline = pipeline(
|
30 |
"automatic-speech-recognition",
|
31 |
model="jonatasgrosman/wav2vec2-large-xlsr-53-arabic",
|
32 |
token=os.getenv("HF_TOKEN")
|
33 |
)
|
34 |
|
35 |
+
# 2. نموذج توليف الصوت الأنثوي (Mishkal TTS)
|
36 |
tts_tokenizer = AutoTokenizer.from_pretrained(
|
37 |
+
"miscellaneous-stuff/mishkal-tts",
|
38 |
token=os.getenv("HF_TOKEN")
|
39 |
)
|
40 |
tts_model = VitsModel.from_pretrained(
|
41 |
+
"miscellaneous-stuff/mishkal-tts",
|
42 |
token=os.getenv("HF_TOKEN")
|
43 |
)
|
44 |
|
|
|
46 |
logger.error(f"فشل تحميل النماذج: {str(e)}")
|
47 |
raise
|
48 |
|
49 |
+
# ===== ذاكرة المحادثة =====
|
50 |
+
conversation_history = defaultdict(list)
|
51 |
+
|
52 |
# ===== دوال معالجة الصوت =====
|
53 |
def enhance_audio(input_path: str, output_path: str) -> bool:
|
54 |
try:
|
55 |
audio = AudioSegment.from_wav(input_path)
|
56 |
+
audio = audio.low_pass_filter(3000) # تقليل الضوضاء
|
57 |
+
audio = audio.high_pass_filter(100) # إزالة الترددات المنخفضة
|
58 |
+
audio = audio.normalize() # توحيد مستوى الصوت
|
59 |
audio = audio.fade_in(150).fade_out(150)
|
60 |
audio.export(output_path, format="wav")
|
61 |
return True
|
|
|
73 |
logger.error(f"فشل التعرف على الصوت: {str(e)}")
|
74 |
return ""
|
75 |
|
76 |
+
async def generate_response(text: str, user_id: str) -> str:
|
77 |
+
"""توليد رد قصير مترابط"""
|
78 |
try:
|
79 |
+
# تحديث ذاكرة المحادثة
|
80 |
+
conversation_history[user_id].append(text)
|
81 |
+
context = "\n".join(conversation_history[user_id][-3:]) # أخر 3 رسائل
|
82 |
+
|
83 |
chatbot = pipeline(
|
84 |
"text-generation",
|
85 |
model="aubmindlab/aragpt2-base",
|
86 |
+
token=os.getenv("HF_TOKEN"),
|
87 |
+
max_length=50, # تقليل طول الرد
|
88 |
+
temperature=0.7, # زيادة التركيز
|
89 |
)
|
90 |
response = chatbot(
|
91 |
+
context,
|
|
|
92 |
num_return_sequences=1,
|
93 |
pad_token_id=50256
|
94 |
)
|
|
|
98 |
return "حدث خطأ في توليد الرد."
|
99 |
|
100 |
async def text_to_speech(text: str) -> None:
|
101 |
+
"""تحويل النص إلى صوت أنثوي"""
|
102 |
try:
|
103 |
inputs = tts_tokenizer(text, return_tensors="pt")
|
104 |
with torch.no_grad():
|
105 |
+
output = tts_model(**inputs, speaker_id=2) # اختيار الصوت الأنثوي
|
106 |
waveform = output.waveform[0].numpy()
|
107 |
sf.write("bot_response.wav", waveform, tts_model.config.sampling_rate)
|
108 |
except Exception as e:
|
109 |
logger.error(f"فشل تحويل النص إلى صوت: {str(e)}")
|
110 |
|
111 |
+
# ===== دوال التفاعل مع المستخدم =====
|
112 |
+
async def start(update: Update, context):
|
113 |
+
"""رسالة الترحيب"""
|
114 |
+
await update.message.reply_text("مرحبًا! أنا بوت الدردشة الصوتية الأنثوي. أرسل لي رسالة صوتية وسأرد عليك بصوت أنثوي عالي الجودة 🎤")
|
115 |
+
|
116 |
async def process_voice(update: Update, context):
|
117 |
try:
|
118 |
+
user_id = update.message.from_user.id
|
119 |
voice_file = await update.message.voice.get_file()
|
120 |
await voice_file.download_to_drive("user_voice.ogg")
|
121 |
|
122 |
+
# معالجة الصوت
|
123 |
user_text = await speech_to_text("user_voice.ogg")
|
124 |
+
bot_response = await generate_response(user_text, str(user_id))
|
125 |
await text_to_speech(bot_response)
|
126 |
|
127 |
+
# إرسال الرد
|
128 |
if enhance_audio("bot_response.wav", "bot_response_enhanced.wav"):
|
129 |
await update.message.reply_voice("bot_response_enhanced.wav")
|
130 |
else:
|
|
|
134 |
logger.error(f"خطأ غير متوقع: {str(e)}")
|
135 |
await update.message.reply_text("⚠️ عذرًا، حدث خطأ في المعالجة.")
|
136 |
|
137 |
+
# ===== التشغيل الرئيسي =====
|
138 |
def run_bot():
|
139 |
loop = asyncio.new_event_loop()
|
140 |
asyncio.set_event_loop(loop)
|
141 |
|
142 |
application = ApplicationBuilder().token(os.getenv("TELEGRAM_TOKEN")).build()
|
143 |
+
application.add_handler(CommandHandler("start", start))
|
144 |
application.add_handler(MessageHandler(filters.VOICE, process_voice))
|
145 |
|
146 |
application.run_polling(
|
|
|
149 |
)
|
150 |
|
151 |
if __name__ == "__main__":
|
|
|
152 |
bot_thread = threading.Thread(target=run_bot, daemon=True)
|
153 |
bot_thread.start()
|
154 |
bot_thread.join()
|