Update app.py
Browse files
app.py
CHANGED
|
@@ -1,4 +1,4 @@
|
|
| 1 |
-
# app.py - نسخه
|
| 2 |
|
| 3 |
import os
|
| 4 |
import sys
|
|
@@ -12,6 +12,8 @@ import logging
|
|
| 12 |
import mimetypes
|
| 13 |
import threading
|
| 14 |
import random
|
|
|
|
|
|
|
| 15 |
from fastapi import FastAPI, HTTPException
|
| 16 |
from pydantic import BaseModel
|
| 17 |
from google import genai
|
|
@@ -36,35 +38,27 @@ def _init_api_keys():
|
|
| 36 |
global ALL_API_KEYS
|
| 37 |
all_keys_string = os.environ.get("ALL_GEMINI_API_KEYS")
|
| 38 |
if all_keys_string:
|
| 39 |
-
# کلیدها را فقط میخوانیم و در لیست نگه میداریم
|
| 40 |
ALL_API_KEYS = [key.strip() for key in all_keys_string.split(',') if key.strip()]
|
| 41 |
logging.info(f"✅ تعداد {len(ALL_API_KEYS)} کلید API جیمینای شناسایی و بارگذاری شد.")
|
| 42 |
if not ALL_API_KEYS:
|
| 43 |
logging.warning("⛔️ هشدار: هیچ Secret با نام ALL_GEMINI_API_KEYS یافت نشد!")
|
| 44 |
|
| 45 |
def get_random_api_key_and_client():
|
| 46 |
-
"""
|
| 47 |
-
یک کلید را به صورت کاملاً تصادفی انتخاب میکند.
|
| 48 |
-
"""
|
| 49 |
if not ALL_API_KEYS:
|
| 50 |
return None, None
|
| 51 |
-
|
| 52 |
-
# انتخاب تصادفی از لیست ثابت
|
| 53 |
key_to_use = random.choice(ALL_API_KEYS)
|
| 54 |
-
|
| 55 |
with CLIENT_CACHE_LOCK:
|
| 56 |
if key_to_use in GEMINI_CLIENTS_CACHE:
|
| 57 |
client = GEMINI_CLIENTS_CACHE[key_to_use]
|
| 58 |
else:
|
| 59 |
-
# ساخت کلاینت جدید و ذخیره در کش
|
| 60 |
client = genai.Client(api_key=key_to_use)
|
| 61 |
GEMINI_CLIENTS_CACHE[key_to_use] = client
|
| 62 |
-
|
| 63 |
return key_to_use, client
|
| 64 |
|
| 65 |
-
|
|
|
|
| 66 |
DEFAULT_MAX_CHUNK_SIZE = 3800
|
| 67 |
-
DEFAULT_SLEEP_BETWEEN_REQUESTS = 5
|
| 68 |
|
| 69 |
def save_binary_file(file_name, data):
|
| 70 |
try:
|
|
@@ -123,82 +117,152 @@ def merge_audio_files_func(file_paths, output_path):
|
|
| 123 |
return True
|
| 124 |
except Exception as e: logging.error(f"❌ خطا در ادغام فایلهای صوتی: {e}"); return False
|
| 125 |
|
| 126 |
-
# --- منطق
|
| 127 |
-
def
|
| 128 |
-
|
|
|
|
|
|
|
|
|
|
| 129 |
|
| 130 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 131 |
|
| 132 |
for attempt in range(MAX_RETRIES):
|
| 133 |
-
|
| 134 |
-
|
| 135 |
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 139 |
|
| 140 |
try:
|
| 141 |
-
|
|
|
|
|
|
|
| 142 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 143 |
final_text = f'{chunk_text}({prompt_text})' if prompt_text and prompt_text.strip() else chunk_text
|
| 144 |
contents = [types.Content(role="user", parts=[types.Part.from_text(text=final_text)])]
|
| 145 |
config = types.GenerateContentConfig(temperature=temp, response_modalities=["audio"],
|
| 146 |
speech_config=types.SpeechConfig(voice_config=types.VoiceConfig(
|
| 147 |
prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name=voice))))
|
| 148 |
|
| 149 |
-
response = client.models.generate_content(model=
|
| 150 |
-
|
| 151 |
if response.candidates and response.candidates[0].content and response.candidates[0].content.parts and response.candidates[0].content.parts[0].inline_data:
|
| 152 |
-
logging.info(f"[{session_id}] ✅ قطعه
|
| 153 |
return response.candidates[0].content.parts[0].inline_data
|
| 154 |
-
|
| 155 |
except Exception as e:
|
| 156 |
-
|
| 157 |
-
logging.warning(f"[{session_id}] ⚠️ خطا در تلاش {attempt+1} (کلید ...{selected_api_key[-4:]}): {e}")
|
| 158 |
-
|
| 159 |
-
# وقفه کوتاه (نیم ثانیه) قبل از انتخاب تصادفی بعدی
|
| 160 |
time.sleep(0.5)
|
| 161 |
-
|
| 162 |
return None
|
| 163 |
|
| 164 |
-
def core_generate_audio(text_input, prompt_input, selected_voice, temperature_val, session_id):
|
| 165 |
-
logging.info(f"[{session_id}] 🚀 شروع
|
| 166 |
temp_dir = f"temp_{session_id}"
|
| 167 |
os.makedirs(temp_dir, exist_ok=True)
|
| 168 |
output_base_name = f"{temp_dir}/audio_session_{session_id}"
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
if not text_chunks: raise ValueError("متن قابل پردازش به قطعات کوچکتر نیست.")
|
| 172 |
-
generated_files = []
|
| 173 |
try:
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
|
| 181 |
-
|
| 182 |
-
|
| 183 |
-
|
| 184 |
-
|
|
|
|
| 185 |
else:
|
| 186 |
-
|
| 187 |
-
|
| 188 |
-
|
| 189 |
-
|
| 190 |
-
|
| 191 |
-
|
| 192 |
-
|
| 193 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 194 |
else:
|
| 195 |
shutil.move(generated_files[0], final_output_path)
|
| 196 |
-
|
| 197 |
-
|
| 198 |
-
|
| 199 |
-
final_audio_file = final_output_path
|
| 200 |
-
logging.info(f"[{session_id}] ✅ فایل صوتی نهایی با موفقیت تولید شد: {os.path.basename(final_audio_file)}")
|
| 201 |
-
return final_audio_file
|
| 202 |
finally:
|
| 203 |
if os.path.exists(temp_dir):
|
| 204 |
shutil.rmtree(temp_dir)
|
|
@@ -212,34 +276,33 @@ class TTSRequest(BaseModel):
|
|
| 212 |
prompt: str | None = ""
|
| 213 |
speaker: str
|
| 214 |
temperature: float
|
|
|
|
| 215 |
|
| 216 |
@app.post("/generate")
|
| 217 |
def generate_audio_endpoint(request: TTSRequest):
|
| 218 |
session_id = str(uuid.uuid4())[:8]
|
| 219 |
-
logging.info(f"[{session_id}] 🏁 درخواست جدید API در این Worker دریافت شد.")
|
| 220 |
try:
|
| 221 |
final_path = core_generate_audio(
|
| 222 |
text_input=request.text,
|
| 223 |
prompt_input=request.prompt,
|
| 224 |
selected_voice=request.speaker,
|
| 225 |
temperature_val=request.temperature,
|
| 226 |
-
session_id=session_id
|
|
|
|
| 227 |
)
|
| 228 |
if final_path and os.path.exists(final_path):
|
| 229 |
from fastapi.responses import FileResponse
|
| 230 |
return FileResponse(path=final_path, media_type='audio/wav', filename=os.path.basename(final_path), background=shutil.rmtree(os.path.dirname(final_path), ignore_errors=True))
|
| 231 |
else:
|
| 232 |
-
raise HTTPException(status_code=500, detail="خطا در تولید فایل صوتی
|
| 233 |
except Exception as e:
|
| 234 |
-
logging.error(f"[{session_id}] ❌ خطا
|
| 235 |
raise HTTPException(status_code=500, detail=str(e))
|
| 236 |
|
| 237 |
@app.get("/")
|
| 238 |
def health_check():
|
| 239 |
return {"status": "ok", "message": "TTS Worker is running."}
|
| 240 |
|
| 241 |
-
logging.info("✅✅✅ Application logic initialized successfully. Starting Uvicorn server...")
|
| 242 |
-
|
| 243 |
if __name__ == "__main__":
|
| 244 |
port = int(os.environ.get("PORT", 7860))
|
| 245 |
uvicorn.run(app, host="0.0.0.0", port=port, reload=False)
|
|
|
|
| 1 |
+
# app.py - نسخه Worker با پشتیبانی از Gemini Live و Standard
|
| 2 |
|
| 3 |
import os
|
| 4 |
import sys
|
|
|
|
| 12 |
import mimetypes
|
| 13 |
import threading
|
| 14 |
import random
|
| 15 |
+
import asyncio
|
| 16 |
+
import wave
|
| 17 |
from fastapi import FastAPI, HTTPException
|
| 18 |
from pydantic import BaseModel
|
| 19 |
from google import genai
|
|
|
|
| 38 |
global ALL_API_KEYS
|
| 39 |
all_keys_string = os.environ.get("ALL_GEMINI_API_KEYS")
|
| 40 |
if all_keys_string:
|
|
|
|
| 41 |
ALL_API_KEYS = [key.strip() for key in all_keys_string.split(',') if key.strip()]
|
| 42 |
logging.info(f"✅ تعداد {len(ALL_API_KEYS)} کلید API جیمینای شناسایی و بارگذاری شد.")
|
| 43 |
if not ALL_API_KEYS:
|
| 44 |
logging.warning("⛔️ هشدار: هیچ Secret با نام ALL_GEMINI_API_KEYS یافت نشد!")
|
| 45 |
|
| 46 |
def get_random_api_key_and_client():
|
|
|
|
|
|
|
|
|
|
| 47 |
if not ALL_API_KEYS:
|
| 48 |
return None, None
|
|
|
|
|
|
|
| 49 |
key_to_use = random.choice(ALL_API_KEYS)
|
|
|
|
| 50 |
with CLIENT_CACHE_LOCK:
|
| 51 |
if key_to_use in GEMINI_CLIENTS_CACHE:
|
| 52 |
client = GEMINI_CLIENTS_CACHE[key_to_use]
|
| 53 |
else:
|
|
|
|
| 54 |
client = genai.Client(api_key=key_to_use)
|
| 55 |
GEMINI_CLIENTS_CACHE[key_to_use] = 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" # مدل لایف
|
| 60 |
DEFAULT_MAX_CHUNK_SIZE = 3800
|
| 61 |
+
DEFAULT_SLEEP_BETWEEN_REQUESTS = 5
|
| 62 |
|
| 63 |
def save_binary_file(file_name, data):
|
| 64 |
try:
|
|
|
|
| 117 |
return True
|
| 118 |
except Exception as e: logging.error(f"❌ خطا در ادغام فایلهای صوتی: {e}"); return False
|
| 119 |
|
| 120 |
+
# --- منطق Gemini Live (جدید) ---
|
| 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(
|
| 131 |
+
voice_config=types.VoiceConfig(
|
| 132 |
+
prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name=voice)
|
| 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) دریافت موفقیتآمیز {len(audio_buffer)} بایت.")
|
| 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) # Mono
|
| 182 |
+
wf.setsampwidth(2) # 16-bit
|
| 183 |
+
wf.setframerate(24000) # 24kHz (استاندارد مدل لایف)
|
| 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 (قدیمی) ---
|
| 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 برای پردازش در دسترس نیست.")
|
| 193 |
+
MAX_RETRIES = 50
|
| 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"],
|
| 201 |
speech_config=types.SpeechConfig(voice_config=types.VoiceConfig(
|
| 202 |
prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name=voice))))
|
| 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) قطعه تولید شد.")
|
| 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 پردازش (Live Mode: {use_live_model})")
|
| 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 |
+
# --- مسیر ۱: استفاده از مدل لای�� ---
|
| 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 |
+
if pcm_data:
|
| 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 پس از تلاشهای مکرر ناموفق بود.")
|
| 235 |
+
|
| 236 |
+
# --- مسیر ۲: استفاده از مدل استاندارد ---
|
| 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 |
+
inline_data = generate_audio_chunk_standard_with_retry(chunk, prompt_input, selected_voice, temperature_val, session_id)
|
| 242 |
+
if inline_data:
|
| 243 |
+
data_buffer = inline_data.data
|
| 244 |
+
ext = mimetypes.guess_extension(inline_data.mime_type) or ".wav"
|
| 245 |
+
if "audio/L" in inline_data.mime_type and ext == ".wav":
|
| 246 |
+
data_buffer = convert_to_wav(data_buffer, inline_data.mime_type)
|
| 247 |
+
if not ext.startswith("."): ext = "." + ext
|
| 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 |
+
raise Exception(f"تولید قطعه {i+1} استاندارد ناموفق بود.")
|
| 252 |
+
if i < len(text_chunks) - 1: time.sleep(DEFAULT_SLEEP_BETWEEN_REQUESTS)
|
| 253 |
+
|
| 254 |
+
if not generated_files: raise Exception("هیچ فایلی تولید نشد.")
|
| 255 |
+
|
| 256 |
+
if len(generated_files) > 1:
|
| 257 |
+
if PYDUB_AVAILABLE and merge_audio_files_func(generated_files, final_output_path):
|
| 258 |
+
pass
|
| 259 |
+
else:
|
| 260 |
+
shutil.move(generated_files[0], final_output_path)
|
| 261 |
else:
|
| 262 |
shutil.move(generated_files[0], final_output_path)
|
| 263 |
+
|
| 264 |
+
return final_output_path
|
| 265 |
+
|
|
|
|
|
|
|
|
|
|
| 266 |
finally:
|
| 267 |
if os.path.exists(temp_dir):
|
| 268 |
shutil.rmtree(temp_dir)
|
|
|
|
| 276 |
prompt: str | None = ""
|
| 277 |
speaker: str
|
| 278 |
temperature: float
|
| 279 |
+
use_live_model: bool = False # پارامتر جدید
|
| 280 |
|
| 281 |
@app.post("/generate")
|
| 282 |
def generate_audio_endpoint(request: TTSRequest):
|
| 283 |
session_id = str(uuid.uuid4())[:8]
|
|
|
|
| 284 |
try:
|
| 285 |
final_path = core_generate_audio(
|
| 286 |
text_input=request.text,
|
| 287 |
prompt_input=request.prompt,
|
| 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
|
| 295 |
return FileResponse(path=final_path, media_type='audio/wav', filename=os.path.basename(final_path), background=shutil.rmtree(os.path.dirname(final_path), ignore_errors=True))
|
| 296 |
else:
|
| 297 |
+
raise HTTPException(status_code=500, detail="خطا در تولید فایل صوتی.")
|
| 298 |
except Exception as e:
|
| 299 |
+
logging.error(f"[{session_id}] ❌ خطا: {e}")
|
| 300 |
raise HTTPException(status_code=500, detail=str(e))
|
| 301 |
|
| 302 |
@app.get("/")
|
| 303 |
def health_check():
|
| 304 |
return {"status": "ok", "message": "TTS Worker is running."}
|
| 305 |
|
|
|
|
|
|
|
| 306 |
if __name__ == "__main__":
|
| 307 |
port = int(os.environ.get("PORT", 7860))
|
| 308 |
uvicorn.run(app, host="0.0.0.0", port=port, reload=False)
|