docker / tts-server.py
maltose1's picture
Upload 4 files
379380b verified
raw
history blame
14.8 kB
import asyncio
import json
import random
import math
import os
import logging
import time
import glob
from typing import Optional, List, Dict, Any, AsyncGenerator
import aiohttp
import websockets
from fastapi import FastAPI, HTTPException, Header, Request, BackgroundTasks, Depends
from fastapi.responses import StreamingResponse, JSONResponse
from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
from pydantic import BaseModel, Field
import uvicorn
from contextlib import asynccontextmanager
# ==========================================
# Configuration & Setup
# ==========================================
# Configure logging
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger("DoubaoTTS")
# Server Settings
PORT = 1547
HOST = "0.0.0.0"
MODELS_FILE = "models.json"
COOKIE_DIR = "cookie" # Directory to store cookie txt files
AUTH_PASSWORD = os.getenv("PASSWORD", "sk-wei123")
# Initialize FastAPI
@asynccontextmanager
async def lifespan(app: FastAPI):
# Load voices from file if exists
if os.path.exists(MODELS_FILE):
try:
with open(MODELS_FILE, "r", encoding="utf-8") as f:
engine.voices = json.load(f)
logger.info(f"Loaded {len(engine.voices)} voices from cache.")
except Exception as e:
logger.error(f"Failed to load cached models: {e}")
# Trigger fetch in background
asyncio.create_task(engine.fetch_voices())
yield
app = FastAPI(title="Doubao TTS OpenAI API Server", lifespan=lifespan)
security = HTTPBearer()
async def verify_token(credentials: HTTPAuthorizationCredentials = Depends(security)):
if credentials.credentials != AUTH_PASSWORD:
raise HTTPException(status_code=401, detail="Invalid authentication token")
return credentials.credentials
# ==========================================
# Cookie Manager
# ==========================================
class CookieManager:
def __init__(self, cookie_dir: str):
self.cookie_dir = cookie_dir
self.cookies: List[str] = []
self.current_index = 0
self.failure_count = 0
self.load_cookies()
def load_cookies(self):
"""Load all .txt files from the cookie directory."""
self.cookies = []
if not os.path.exists(self.cookie_dir):
try:
os.makedirs(self.cookie_dir, exist_ok=True)
logger.info(f"Created cookie directory: {self.cookie_dir}")
except Exception as e:
logger.error(f"Failed to create cookie directory: {e}")
return
files = glob.glob(os.path.join(self.cookie_dir, "*.txt"))
files.sort() # Ensure deterministic order
for f_path in files:
try:
with open(f_path, 'r', encoding='utf-8') as f:
content = f.read().strip()
if content:
self.cookies.append(content)
except Exception as e:
logger.error(f"Error reading cookie file {f_path}: {e}")
logger.info(f"Loaded {len(self.cookies)} cookies from {self.cookie_dir}")
def get_cookie(self) -> Optional[str]:
"""Get the current active cookie."""
if not self.cookies:
# Try reloading in case user added files
self.load_cookies()
if not self.cookies:
return None
# Safety check index
if self.current_index >= len(self.cookies):
self.current_index = 0
return self.cookies[self.current_index]
def report_failure(self):
"""Report a failure for the current cookie. Switches if fails >= 2 times."""
self.failure_count += 1
logger.warning(f"Cookie index {self.current_index} failed. Count: {self.failure_count}/2")
if self.failure_count >= 2:
self._rotate()
def report_success(self):
"""Reset failure count on success."""
if self.failure_count > 0:
self.failure_count = 0
def _rotate(self):
"""Switch to the next cookie."""
if not self.cookies:
return
prev_index = self.current_index
self.current_index = (self.current_index + 1) % len(self.cookies)
self.failure_count = 0
logger.info(f"Rotating cookie: {prev_index} -> {self.current_index}")
# ==========================================
# Core Logic: Doubao Client
# ==========================================
class DoubaoTTS:
def __init__(self):
self.ws_url = "wss://ws-samantha.doubao.com/samantha/audio/tts"
self.voice_url = "https://www.doubao.com/alice/user_voice/recommend"
self.voices = []
self.req_count = 0
self.current_id = self._generate_id()
self.cookie_manager = CookieManager(COOKIE_DIR)
def _generate_id(self) -> str:
"""Generate a random device/session ID."""
num1 = math.floor(1e8 + 9e8 * random.random())
num2 = math.floor(1e8 + 9e8 * random.random())
return f"{num1}{num2}"
def _get_common_params(self) -> str:
"""Generate the common query parameters for requests."""
self.req_count += 1
if self.req_count > 5:
self.req_count = 0
self.current_id = self._generate_id()
cid = self.current_id
return (
f"&mode=0&language=zh&browser_language=zh-CN&device_platform=web"
f"&aid=586861&real_aid=586861&pkg_type=release_version"
f"&device_id={cid}&tea_uuid={cid}&web_id={cid}"
f"&is_new_user=0&region=CN&sys_region=CN"
f"&use-olympus-account=1&samantha_web=1"
f"&version=1.20.1&version_code=20800&pc_version=2.47.2"
)
async def fetch_voices(self):
"""Fetch available voices from Doubao and save to models.json."""
cookie = self.cookie_manager.get_cookie()
if not cookie:
logger.error("Cannot fetch voices: No cookies available.")
return
headers = {
"Cookie": cookie,
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/133.0.0.0 Safari/537.36 Edg/133.0.0.0"
}
# Categories to fetch
tabs = [
(10, "female"),
(10, "male"),
(10, "characters"),
(10, "accent")
]
all_voices_map = {}
async with aiohttp.ClientSession() as session:
for rec_type, tab_key in tabs:
params = self._get_common_params().strip('&')
url = f"{self.voice_url}?{params}"
payload = {
"page_index": 1,
"page_size": 200,
"recommend_type": rec_type,
"tab_key": tab_key
}
try:
async with session.post(url, json=payload, headers=headers) as resp:
if resp.status == 200:
data = await resp.json()
if data.get("code") == 0:
v_list = data.get("data", {}).get("ugc_voice_list", [])
for v in v_list:
vid = v.get("style_id")
if vid and vid not in all_voices_map:
# Construct OpenAI-compatible model object
tag_list = v.get("tag_list", [])
tags = "|".join([t.get("tag_value", "") for t in tag_list])
name = f"{v.get('name')} {tags}".strip()
all_voices_map[vid] = {
"id": vid,
"object": "model",
"created": int(time.time()),
"owned_by": "doubao",
"name": name,
"language": v.get("language_code"),
"icon": v.get("icon", {}).get("url")
}
logger.info(f"Fetched {len(v_list)} voices for tab '{tab_key}'")
else:
logger.warning(f"API Error for tab '{tab_key}': {data.get('msg')}")
else:
logger.warning(f"HTTP {resp.status} for tab '{tab_key}'")
except Exception as e:
logger.error(f"Exception fetching voices for tab '{tab_key}': {e}")
self.voices = list(all_voices_map.values())
# Save to file
try:
with open(MODELS_FILE, "w", encoding="utf-8") as f:
json.dump(self.voices, f, ensure_ascii=False, indent=2)
logger.info(f"Successfully saved {len(self.voices)} voices to {MODELS_FILE}")
except Exception as e:
logger.error(f"Failed to save models.json: {e}")
async def stream_audio(self, text: str, voice: str, speed: float = 1.0, pitch: float = 1.0) -> AsyncGenerator[bytes, None]:
"""Connect to WebSocket and yield audio chunks with retry logic."""
# Map OpenAI speed (0.25 - 4.0) to Doubao rate (-100 to 100)
doubao_rate = int((speed - 1) * 100)
doubao_rate = max(-100, min(100, doubao_rate))
doubao_pitch = int(pitch)
max_retries = 3 # Try up to 3 different cookies/attempts
for attempt in range(max_retries):
cookie = self.cookie_manager.get_cookie()
if not cookie:
logger.error("No cookies available for streaming.")
yield b"" # End stream
return
params = self._get_common_params()
ws_url = f"{self.ws_url}?format=aac&speaker={voice}&speech_rate={doubao_rate}&pitch={doubao_pitch}{params}"
headers = {
"Cookie": cookie,
"Origin": "chrome-extension://capohkkfagimodmlpnahjoijgoocdjhd",
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/133.0.0.0 Safari/537.36 Edg/133.0.0.0"
}
try:
# Try to use extra_headers, if it fails, it might be a version issue or specific environment quirk
# But standard websockets library uses extra_headers.
# Let's try to be safe and use the standard way.
async with websockets.connect(ws_url, extra_headers=headers) as ws:
# Send Text Event
msg = {
"event": "text",
"podcast_extra": {"role": ""},
"text": text
}
await ws.send(json.dumps(msg))
await ws.send(json.dumps({"event": "finish"}))
first_chunk_received = False
async for message in ws:
if isinstance(message, bytes):
if not first_chunk_received:
first_chunk_received = True
self.cookie_manager.report_success() # Mark cookie as good
yield message
elif isinstance(message, str):
# Check for error messages in text frames
# Sometimes errors come as text JSON
pass
# If we finished the loop naturally, break retry loop
return
except websockets.exceptions.InvalidStatusCode as e:
logger.warning(f"WebSocket Handshake failed (Attempt {attempt+1}): {e.status_code}")
self.cookie_manager.report_failure()
# Continue to next attempt (which will get a new cookie if rotated)
except Exception as e:
logger.error(f"WebSocket error (Attempt {attempt+1}): {e}")
self.cookie_manager.report_failure()
# Continue to next attempt
logger.error("All retry attempts failed.")
raise HTTPException(status_code=500, detail="Failed to generate audio after retries.")
# Initialize Engine
engine = DoubaoTTS()
# ==========================================
# API Models
# ==========================================
class OpenAIRequest(BaseModel):
model: Optional[str] = "tts-1"
input: str
voice: str
response_format: Optional[str] = "aac"
speed: Optional[float] = 1.0
pitch: Optional[float] = 0.0
# ==========================================
# API Endpoints
# ==========================================
@app.get("/")
async def root():
return {"status": "running", "service": "Doubao TTS OpenAI Server", "cookies_loaded": len(engine.cookie_manager.cookies)}
@app.get("/v1/models")
async def get_models():
return {"object": "list", "data": engine.voices}
@app.get("/v1/audio/speech")
async def check_speech_endpoint():
"""
Allow GET requests for connectivity checks.
"""
return {"status": "ok", "message": "Speech endpoint is ready"}
@app.post("/v1/audio/speech")
async def create_speech(req: OpenAIRequest, token: str = Depends(verify_token)):
"""
OpenAI-compatible speech generation endpoint.
"""
if not req.input:
raise HTTPException(status_code=400, detail="Input text is required")
media_type = "audio/aac"
if req.response_format == "mp3":
media_type = "audio/mpeg"
return StreamingResponse(
engine.stream_audio(req.input, req.voice, req.speed, req.pitch),
media_type=media_type
)
if __name__ == "__main__":
print(f"Starting server on port {PORT}...")
print(f"Please place cookie files in '{COOKIE_DIR}/' directory (e.g. 1.txt, 2.txt)")
uvicorn.run(app, host=HOST, port=PORT)