Update app.py
Browse files
app.py
CHANGED
|
@@ -1,4 +1,4 @@
|
|
| 1 |
-
# app.py - نسخه Worker با پشتیبانی از
|
| 2 |
|
| 3 |
import os
|
| 4 |
import sys
|
|
@@ -56,7 +56,7 @@ def get_random_api_key_and_client():
|
|
| 56 |
return key_to_use, client
|
| 57 |
|
| 58 |
FIXED_MODEL_NAME_STANDARD = "gemini-2.5-flash-preview-tts"
|
| 59 |
-
FIXED_MODEL_NAME_LIVE = "models/gemini-2.5-flash-native-audio-preview-12-2025"
|
| 60 |
DEFAULT_MAX_CHUNK_SIZE = 3800
|
| 61 |
DEFAULT_SLEEP_BETWEEN_REQUESTS = 5
|
| 62 |
|
|
@@ -117,14 +117,9 @@ def merge_audio_files_func(file_paths, output_path):
|
|
| 117 |
return True
|
| 118 |
except Exception as e: logging.error(f"❌ خطا در ادغام فایلهای صوتی: {e}"); return False
|
| 119 |
|
| 120 |
-
# --- منطق Gemini Live
|
| 121 |
async def generate_audio_live_with_retry(text, prompt, voice, session_id):
|
| 122 |
-
|
| 123 |
-
اتصال به مدل لایف با استفاده از وبسوکت و دریافت صدا.
|
| 124 |
-
"""
|
| 125 |
-
MAX_RETRIES = 50
|
| 126 |
-
|
| 127 |
-
# تنظیمات مدل لایف
|
| 128 |
live_config = types.LiveConnectConfig(
|
| 129 |
response_modalities=["AUDIO"],
|
| 130 |
speech_config=types.SpeechConfig(
|
|
@@ -133,68 +128,53 @@ async def generate_audio_live_with_retry(text, prompt, voice, session_id):
|
|
| 133 |
)
|
| 134 |
),
|
| 135 |
)
|
| 136 |
-
|
| 137 |
for attempt in range(MAX_RETRIES):
|
| 138 |
selected_api_key, _ = get_random_api_key_and_client()
|
| 139 |
if not selected_api_key: break
|
| 140 |
-
|
| 141 |
-
# برای لایف نیاز به کلاینت Async جدید داریم که با کلید خاص ساخته شود
|
| 142 |
-
# چون کلاینتهای کش شده ممکن است سینک باشند یا تنظیمات متفاوتی داشته باشند
|
| 143 |
client = genai.Client(http_options={"api_version": "v1beta"}, api_key=selected_api_key)
|
| 144 |
-
|
| 145 |
unique_id_for_req = str(uuid.uuid4())[:8]
|
| 146 |
tts_prompt = f"Please read the following text naturally: '{text}' [ID: {unique_id_for_req}]"
|
| 147 |
-
if prompt:
|
| 148 |
-
tts_prompt = f"With a {prompt} tone, please read: '{text}'"
|
| 149 |
-
|
| 150 |
try:
|
| 151 |
logging.info(f"[{session_id}] (Live) تلاش {attempt+1} با کلید ...{selected_api_key[-4:]}")
|
| 152 |
-
|
| 153 |
audio_buffer = bytearray()
|
| 154 |
-
|
| 155 |
async with client.aio.live.connect(model=FIXED_MODEL_NAME_LIVE, config=live_config) as session:
|
| 156 |
await session.send(input=tts_prompt, end_of_turn=True)
|
| 157 |
-
|
| 158 |
-
# دریافت استریم
|
| 159 |
async for response in session.receive():
|
| 160 |
-
if response.data:
|
| 161 |
-
audio_buffer.extend(response.data)
|
| 162 |
-
if response.text:
|
| 163 |
-
pass # متن را نادیده میگیریم
|
| 164 |
-
|
| 165 |
if len(audio_buffer) > 0:
|
| 166 |
-
logging.info(f"[{session_id}] ✅ (Live)
|
| 167 |
return audio_buffer
|
| 168 |
-
else:
|
| 169 |
-
raise Exception("بافر صوتی خالی بود.")
|
| 170 |
-
|
| 171 |
except Exception as e:
|
| 172 |
logging.warning(f"[{session_id}] ⚠️ (Live) خطا در تلاش {attempt+1}: {e}")
|
| 173 |
time.sleep(0.5)
|
| 174 |
-
|
| 175 |
return None
|
| 176 |
|
| 177 |
def save_pcm_to_wav(pcm_data, output_path):
|
| 178 |
-
"""ذخیره دیتای خام PCM مدل لایف به فرمت WAV استاندارد"""
|
| 179 |
try:
|
| 180 |
with wave.open(output_path, 'wb') as wf:
|
| 181 |
-
wf.setnchannels(1)
|
| 182 |
-
wf.setsampwidth(2)
|
| 183 |
-
wf.setframerate(24000)
|
| 184 |
wf.writeframes(pcm_data)
|
| 185 |
return True
|
| 186 |
except Exception as e:
|
| 187 |
logging.error(f"خطا در تبدیل PCM به WAV: {e}")
|
| 188 |
return False
|
| 189 |
|
| 190 |
-
# --- منطق Gemini Standard (
|
| 191 |
-
def generate_audio_chunk_standard_with_retry(chunk_text, prompt_text, voice, temp, session_id):
|
| 192 |
-
if not ALL_API_KEYS: raise Exception("هیچ کلید API
|
| 193 |
-
|
|
|
|
|
|
|
|
|
|
| 194 |
for attempt in range(MAX_RETRIES):
|
| 195 |
selected_api_key, client = get_random_api_key_and_client()
|
| 196 |
if not client: break
|
| 197 |
try:
|
|
|
|
| 198 |
final_text = f'{chunk_text}({prompt_text})' if prompt_text and prompt_text.strip() else chunk_text
|
| 199 |
contents = [types.Content(role="user", parts=[types.Part.from_text(text=final_text)])]
|
| 200 |
config = types.GenerateContentConfig(temperature=temp, response_modalities=["audio"],
|
|
@@ -203,42 +183,39 @@ def generate_audio_chunk_standard_with_retry(chunk_text, prompt_text, voice, tem
|
|
| 203 |
|
| 204 |
response = client.models.generate_content(model=FIXED_MODEL_NAME_STANDARD, contents=contents, config=config)
|
| 205 |
if response.candidates and response.candidates[0].content and response.candidates[0].content.parts and response.candidates[0].content.parts[0].inline_data:
|
| 206 |
-
logging.info(f"[{session_id}] ✅ (Standard)
|
| 207 |
return response.candidates[0].content.parts[0].inline_data
|
| 208 |
except Exception as e:
|
| 209 |
logging.warning(f"[{session_id}] ⚠️ (Standard) خطا در تلاش {attempt+1}: {e}")
|
| 210 |
time.sleep(0.5)
|
| 211 |
return None
|
| 212 |
|
| 213 |
-
def core_generate_audio(text_input, prompt_input, selected_voice, temperature_val, session_id, use_live_model=False):
|
| 214 |
-
logging.info(f"[{session_id}] 🚀 شروع
|
| 215 |
temp_dir = f"temp_{session_id}"
|
| 216 |
os.makedirs(temp_dir, exist_ok=True)
|
| 217 |
output_base_name = f"{temp_dir}/audio_session_{session_id}"
|
| 218 |
final_output_path = f"output_{session_id}.wav"
|
| 219 |
|
| 220 |
try:
|
| 221 |
-
#
|
| 222 |
if use_live_model:
|
| 223 |
-
# در مدل لایف، متن را تکه تکه نمیکنیم (چون شرط <500 کاراکتر چک شده)
|
| 224 |
-
# باید تابع async را در محیط sync اجرا کنیم
|
| 225 |
pcm_data = asyncio.run(generate_audio_live_with_retry(text_input, prompt_input, selected_voice, session_id))
|
| 226 |
-
|
| 227 |
-
|
| 228 |
-
if save_pcm_to_wav(pcm_data, final_output_path):
|
| 229 |
-
logging.info(f"[{session_id}] ✅ فایل لایف ذخیره شد.")
|
| 230 |
-
return final_output_path
|
| 231 |
-
else:
|
| 232 |
-
raise Exception("خطا در ذخیره فایل WAV لایف.")
|
| 233 |
else:
|
| 234 |
-
raise Exception("تولید صدا با مدل لایف
|
| 235 |
|
| 236 |
-
#
|
| 237 |
else:
|
| 238 |
text_chunks = smart_text_split(text_input, DEFAULT_MAX_CHUNK_SIZE)
|
| 239 |
generated_files = []
|
|
|
|
|
|
|
| 240 |
for i, chunk in enumerate(text_chunks):
|
| 241 |
-
|
|
|
|
|
|
|
| 242 |
if inline_data:
|
| 243 |
data_buffer = inline_data.data
|
| 244 |
ext = mimetypes.guess_extension(inline_data.mime_type) or ".wav"
|
|
@@ -248,9 +225,25 @@ def core_generate_audio(text_input, prompt_input, selected_voice, temperature_va
|
|
| 248 |
fpath = save_binary_file(f"{output_base_name}_part{i+1:03d}{ext}", data_buffer)
|
| 249 |
if fpath: generated_files.append(fpath)
|
| 250 |
else:
|
| 251 |
-
|
| 252 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 253 |
|
|
|
|
| 254 |
if not generated_files: raise Exception("هیچ فایلی تولید نشد.")
|
| 255 |
|
| 256 |
if len(generated_files) > 1:
|
|
@@ -276,7 +269,9 @@ class TTSRequest(BaseModel):
|
|
| 276 |
prompt: str | None = ""
|
| 277 |
speaker: str
|
| 278 |
temperature: float
|
| 279 |
-
use_live_model: bool = False
|
|
|
|
|
|
|
| 280 |
|
| 281 |
@app.post("/generate")
|
| 282 |
def generate_audio_endpoint(request: TTSRequest):
|
|
@@ -288,7 +283,9 @@ def generate_audio_endpoint(request: TTSRequest):
|
|
| 288 |
selected_voice=request.speaker,
|
| 289 |
temperature_val=request.temperature,
|
| 290 |
session_id=session_id,
|
| 291 |
-
use_live_model=request.use_live_model
|
|
|
|
|
|
|
| 292 |
)
|
| 293 |
if final_path and os.path.exists(final_path):
|
| 294 |
from fastapi.responses import FileResponse
|
|
|
|
| 1 |
+
# app.py - نسخه Worker با پشتیبانی از Fallback و Retry Limit
|
| 2 |
|
| 3 |
import os
|
| 4 |
import sys
|
|
|
|
| 56 |
return key_to_use, client
|
| 57 |
|
| 58 |
FIXED_MODEL_NAME_STANDARD = "gemini-2.5-flash-preview-tts"
|
| 59 |
+
FIXED_MODEL_NAME_LIVE = "models/gemini-2.5-flash-native-audio-preview-12-2025"
|
| 60 |
DEFAULT_MAX_CHUNK_SIZE = 3800
|
| 61 |
DEFAULT_SLEEP_BETWEEN_REQUESTS = 5
|
| 62 |
|
|
|
|
| 117 |
return True
|
| 118 |
except Exception as e: logging.error(f"❌ خطا در ادغام فایلهای صوتی: {e}"); return False
|
| 119 |
|
| 120 |
+
# --- منطق Gemini Live ---
|
| 121 |
async def generate_audio_live_with_retry(text, prompt, voice, session_id):
|
| 122 |
+
MAX_RETRIES = 50
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 123 |
live_config = types.LiveConnectConfig(
|
| 124 |
response_modalities=["AUDIO"],
|
| 125 |
speech_config=types.SpeechConfig(
|
|
|
|
| 128 |
)
|
| 129 |
),
|
| 130 |
)
|
|
|
|
| 131 |
for attempt in range(MAX_RETRIES):
|
| 132 |
selected_api_key, _ = get_random_api_key_and_client()
|
| 133 |
if not selected_api_key: break
|
|
|
|
|
|
|
|
|
|
| 134 |
client = genai.Client(http_options={"api_version": "v1beta"}, api_key=selected_api_key)
|
|
|
|
| 135 |
unique_id_for_req = str(uuid.uuid4())[:8]
|
| 136 |
tts_prompt = f"Please read the following text naturally: '{text}' [ID: {unique_id_for_req}]"
|
| 137 |
+
if prompt: tts_prompt = f"With a {prompt} tone, please read: '{text}'"
|
|
|
|
|
|
|
| 138 |
try:
|
| 139 |
logging.info(f"[{session_id}] (Live) تلاش {attempt+1} با کلید ...{selected_api_key[-4:]}")
|
|
|
|
| 140 |
audio_buffer = bytearray()
|
|
|
|
| 141 |
async with client.aio.live.connect(model=FIXED_MODEL_NAME_LIVE, config=live_config) as session:
|
| 142 |
await session.send(input=tts_prompt, end_of_turn=True)
|
|
|
|
|
|
|
| 143 |
async for response in session.receive():
|
| 144 |
+
if response.data: audio_buffer.extend(response.data)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 145 |
if len(audio_buffer) > 0:
|
| 146 |
+
logging.info(f"[{session_id}] ✅ (Live) موفقیتآمیز.")
|
| 147 |
return audio_buffer
|
| 148 |
+
else: raise Exception("بافر صوتی خالی بود.")
|
|
|
|
|
|
|
| 149 |
except Exception as e:
|
| 150 |
logging.warning(f"[{session_id}] ⚠️ (Live) خطا در تلاش {attempt+1}: {e}")
|
| 151 |
time.sleep(0.5)
|
|
|
|
| 152 |
return None
|
| 153 |
|
| 154 |
def save_pcm_to_wav(pcm_data, output_path):
|
|
|
|
| 155 |
try:
|
| 156 |
with wave.open(output_path, 'wb') as wf:
|
| 157 |
+
wf.setnchannels(1)
|
| 158 |
+
wf.setsampwidth(2)
|
| 159 |
+
wf.setframerate(24000)
|
| 160 |
wf.writeframes(pcm_data)
|
| 161 |
return True
|
| 162 |
except Exception as e:
|
| 163 |
logging.error(f"خطا در تبدیل PCM به WAV: {e}")
|
| 164 |
return False
|
| 165 |
|
| 166 |
+
# --- منطق Gemini Standard (اصلاح شده با retry_limit) ---
|
| 167 |
+
def generate_audio_chunk_standard_with_retry(chunk_text, prompt_text, voice, temp, session_id, retry_limit):
|
| 168 |
+
if not ALL_API_KEYS: raise Exception("هیچ کلید API در دسترس نیست.")
|
| 169 |
+
|
| 170 |
+
# استفاده از محدودیت تعیین شده توسط Manager
|
| 171 |
+
MAX_RETRIES = retry_limit
|
| 172 |
+
|
| 173 |
for attempt in range(MAX_RETRIES):
|
| 174 |
selected_api_key, client = get_random_api_key_and_client()
|
| 175 |
if not client: break
|
| 176 |
try:
|
| 177 |
+
# logging.info(f"[{session_id}] (Standard) تلاش {attempt+1}/{MAX_RETRIES} با کلید ...{selected_api_key[-4:]}")
|
| 178 |
final_text = f'{chunk_text}({prompt_text})' if prompt_text and prompt_text.strip() else chunk_text
|
| 179 |
contents = [types.Content(role="user", parts=[types.Part.from_text(text=final_text)])]
|
| 180 |
config = types.GenerateContentConfig(temperature=temp, response_modalities=["audio"],
|
|
|
|
| 183 |
|
| 184 |
response = client.models.generate_content(model=FIXED_MODEL_NAME_STANDARD, contents=contents, config=config)
|
| 185 |
if response.candidates and response.candidates[0].content and response.candidates[0].content.parts and response.candidates[0].content.parts[0].inline_data:
|
| 186 |
+
logging.info(f"[{session_id}] ✅ (Standard) موفقیت در تلاش {attempt+1}.")
|
| 187 |
return response.candidates[0].content.parts[0].inline_data
|
| 188 |
except Exception as e:
|
| 189 |
logging.warning(f"[{session_id}] ⚠️ (Standard) خطا در تلاش {attempt+1}: {e}")
|
| 190 |
time.sleep(0.5)
|
| 191 |
return None
|
| 192 |
|
| 193 |
+
def core_generate_audio(text_input, prompt_input, selected_voice, temperature_val, session_id, use_live_model=False, retry_limit=50, fallback_to_live=False):
|
| 194 |
+
logging.info(f"[{session_id}] 🚀 شروع: Live={use_live_model}, Retry={retry_limit}, Fallback={fallback_to_live}")
|
| 195 |
temp_dir = f"temp_{session_id}"
|
| 196 |
os.makedirs(temp_dir, exist_ok=True)
|
| 197 |
output_base_name = f"{temp_dir}/audio_session_{session_id}"
|
| 198 |
final_output_path = f"output_{session_id}.wav"
|
| 199 |
|
| 200 |
try:
|
| 201 |
+
# 1. اگر دستور مستقیم استفاده از لایف باشد (مثلاً کاربر رایگان)
|
| 202 |
if use_live_model:
|
|
|
|
|
|
|
| 203 |
pcm_data = asyncio.run(generate_audio_live_with_retry(text_input, prompt_input, selected_voice, session_id))
|
| 204 |
+
if pcm_data and save_pcm_to_wav(pcm_data, final_output_path):
|
| 205 |
+
return final_output_path
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 206 |
else:
|
| 207 |
+
raise Exception("تولید صدا با مدل لایف ناموفق بود.")
|
| 208 |
|
| 209 |
+
# 2. استفاده از مدل استاندارد
|
| 210 |
else:
|
| 211 |
text_chunks = smart_text_split(text_input, DEFAULT_MAX_CHUNK_SIZE)
|
| 212 |
generated_files = []
|
| 213 |
+
standard_failed = False
|
| 214 |
+
|
| 215 |
for i, chunk in enumerate(text_chunks):
|
| 216 |
+
# تلاش با مدل استاندارد به تعداد retry_limit
|
| 217 |
+
inline_data = generate_audio_chunk_standard_with_retry(chunk, prompt_input, selected_voice, temperature_val, session_id, retry_limit)
|
| 218 |
+
|
| 219 |
if inline_data:
|
| 220 |
data_buffer = inline_data.data
|
| 221 |
ext = mimetypes.guess_extension(inline_data.mime_type) or ".wav"
|
|
|
|
| 225 |
fpath = save_binary_file(f"{output_base_name}_part{i+1:03d}{ext}", data_buffer)
|
| 226 |
if fpath: generated_files.append(fpath)
|
| 227 |
else:
|
| 228 |
+
standard_failed = True
|
| 229 |
+
break # شکست در تولید یکی از چانکها
|
| 230 |
+
|
| 231 |
+
# 3. بررسی شکست و Fallback
|
| 232 |
+
if standard_failed:
|
| 233 |
+
if fallback_to_live:
|
| 234 |
+
logging.info(f"[{session_id}] 🔄 مدل استاندارد شک��ت خورد. سوییچ به مدل لایف (Fallback)...")
|
| 235 |
+
# پاکسازی فایلهای ناقص قبلی
|
| 236 |
+
generated_files = []
|
| 237 |
+
# فراخوانی مدل لایف برای کل متن
|
| 238 |
+
pcm_data = asyncio.run(generate_audio_live_with_retry(text_input, prompt_input, selected_voice, session_id))
|
| 239 |
+
if pcm_data and save_pcm_to_wav(pcm_data, final_output_path):
|
| 240 |
+
return final_output_path
|
| 241 |
+
else:
|
| 242 |
+
raise Exception("هم مدل استاندارد و هم مدل لایف (Fallback) شکست خوردند.")
|
| 243 |
+
else:
|
| 244 |
+
raise Exception(f"تولید صدا با مدل استاندارد پس از {retry_limit} تلاش ناموفق بود.")
|
| 245 |
|
| 246 |
+
# اگر استاندارد موفق بود، فایلها را ادغام کن
|
| 247 |
if not generated_files: raise Exception("هیچ فایلی تولید نشد.")
|
| 248 |
|
| 249 |
if len(generated_files) > 1:
|
|
|
|
| 269 |
prompt: str | None = ""
|
| 270 |
speaker: str
|
| 271 |
temperature: float
|
| 272 |
+
use_live_model: bool = False
|
| 273 |
+
retry_limit: int = 50 # پارامتر جدید
|
| 274 |
+
fallback_to_live: bool = False # پارامتر جدید
|
| 275 |
|
| 276 |
@app.post("/generate")
|
| 277 |
def generate_audio_endpoint(request: TTSRequest):
|
|
|
|
| 283 |
selected_voice=request.speaker,
|
| 284 |
temperature_val=request.temperature,
|
| 285 |
session_id=session_id,
|
| 286 |
+
use_live_model=request.use_live_model,
|
| 287 |
+
retry_limit=request.retry_limit,
|
| 288 |
+
fallback_to_live=request.fallback_to_live
|
| 289 |
)
|
| 290 |
if final_path and os.path.exists(final_path):
|
| 291 |
from fastapi.responses import FileResponse
|