bdim/Canto
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

feat: add admin usage statistics and LLM configuration management

Browse Source
This commit is contained in:
bdim404
2026-03-12 16:30:24 +08:00
parent 202f2fa83b
commit 7f25dd09f6
16 changed files with 757 additions and 300 deletions
Download Patch File Download Diff File Expand all files Collapse all files

118
qwen3-tts-backend/core/audiobook_service.py
View File

@@ -26,14 +26,18 @@ def cancel_batch(project_id: int) -> None:
logger.info(f"cancel_batch: project={project_id} cancellation signalled")
def _get_llm_service(user: User) -> LLMService:
def _get_llm_service(db: Session) -> LLMService:
from core.security import decrypt_api_key
if not user.llm_api_key or not user.llm_base_url or not user.llm_model:
raise ValueError("LLM config not set. Please configure LLM API key, base URL, and model.")
api_key = decrypt_api_key(user.llm_api_key)
from db.crud import get_system_setting
api_key_encrypted = get_system_setting(db, "llm_api_key")
base_url = get_system_setting(db, "llm_base_url")
model = get_system_setting(db, "llm_model")
if not api_key_encrypted or not base_url or not model:
raise ValueError("LLM config not set. Please configure LLM API key, base URL, and model in admin settings.")
api_key = decrypt_api_key(api_key_encrypted)
if not api_key:
raise ValueError("Failed to decrypt LLM API key.")
return LLMService(base_url=user.llm_base_url, api_key=api_key, model=user.llm_model)
return LLMService(base_url=base_url, api_key=api_key, model=model)
def _get_gendered_instruct(gender: Optional[str], base_instruct: str) -> str:
@@ -167,7 +171,18 @@ async def analyze_project(project_id: int, user: User, db: Session, turbo: bool
crud.update_audiobook_project_status(db, project_id, "analyzing")
ps.append_line(key, f"[分析] 项目「{project.title}」开始角色分析")
llm = _get_llm_service(user)
llm = _get_llm_service(db)
_llm_model = crud.get_system_setting(db, "llm_model")
_user_id = user.id
def _log_analyze_usage(prompt_tokens: int, completion_tokens: int) -> None:
from db.database import SessionLocal
log_db = SessionLocal()
try:
crud.create_usage_log(log_db, _user_id, prompt_tokens, completion_tokens,
model=_llm_model, context="audiobook_analyze")
finally:
log_db.close()
if project.source_type == "epub" and project.source_path:
ps.append_line(key, "[解析] 正在提取 EPUB 章节内容...")
@@ -219,6 +234,7 @@ async def analyze_project(project_id: int, user: User, db: Session, turbo: bool
on_token=on_token,
on_sample=on_sample,
turbo=turbo,
usage_callback=_log_analyze_usage,
)
has_narrator = any(c.get("name") == "narrator" for c in characters_data)
@@ -356,7 +372,19 @@ async def parse_one_chapter(project_id: int, chapter_id: int, user: User, db) ->
try:
crud.update_audiobook_chapter_status(db, chapter_id, "parsing")
llm = _get_llm_service(user)
llm = _get_llm_service(db)
_llm_model = crud.get_system_setting(db, "llm_model")
_user_id = user.id
def _log_parse_usage(prompt_tokens: int, completion_tokens: int) -> None:
from db.database import SessionLocal
log_db = SessionLocal()
try:
crud.create_usage_log(log_db, _user_id, prompt_tokens, completion_tokens,
model=_llm_model, context="audiobook_parse")
finally:
log_db.close()
characters = crud.list_audiobook_characters(db, project_id)
if not characters:
raise ValueError("No characters found. Please analyze the project first.")
@@ -383,7 +411,7 @@ async def parse_one_chapter(project_id: int, chapter_id: int, user: User, db) ->
ps.append_token(key, token)
try:
segments_data = await llm.parse_chapter_segments(chunk, character_names, on_token=on_token)
segments_data = await llm.parse_chapter_segments(chunk, character_names, on_token=on_token, usage_callback=_log_parse_usage)
except Exception as e:
logger.warning(f"Chapter {chapter_id} chunk {i} failed: {e}")
ps.append_line(key, f"\n[回退] {e}")
@@ -543,8 +571,11 @@ async def generate_project(project_id: int, user: User, db: Session, chapter_ind
backend_type = user.user_preferences.get("default_backend", "aliyun") if user.user_preferences else "aliyun"
user_api_key = None
if backend_type == "aliyun" and user.aliyun_api_key:
user_api_key = decrypt_api_key(user.aliyun_api_key)
if backend_type == "aliyun":
from db.crud import get_system_setting
encrypted = get_system_setting(db, "aliyun_api_key")
if encrypted:
user_api_key = decrypt_api_key(encrypted)
backend = await TTSServiceFactory.get_backend(backend_type, user_api_key)
@@ -572,7 +603,23 @@ async def generate_project(project_id: int, user: User, db: Session, chapter_ind
audio_filename = f"ch{seg.chapter_index:03d}_seg{seg.segment_index:04d}.wav"
audio_path = output_base / audio_filename
if backend_type == "aliyun":
ref_audio_for_emo = design.ref_audio_path
if not ref_audio_for_emo:
preview_path = Path(settings.OUTPUT_DIR) / "audiobook" / str(project_id) / "previews" / f"char_{char.id}.wav"
if preview_path.exists():
ref_audio_for_emo = str(preview_path)
if seg.emo_text and ref_audio_for_emo and Path(ref_audio_for_emo).exists():
from core.tts_service import IndexTTS2Backend
indextts2 = IndexTTS2Backend()
audio_bytes = await indextts2.generate(
text=seg.text,
spk_audio_prompt=ref_audio_for_emo,
output_path=str(audio_path),
emo_text=seg.emo_text,
emo_alpha=seg.emo_alpha if seg.emo_alpha is not None else 0.6,
)
elif backend_type == "aliyun":
if design.aliyun_voice_id:
audio_bytes, _ = await backend.generate_voice_design(
{"text": seg.text, "language": "zh"},
@@ -584,16 +631,6 @@ async def generate_project(project_id: int, user: User, db: Session, chapter_ind
"language": "zh",
"instruct": _get_gendered_instruct(char.gender, design.instruct),
})
elif char.use_indextts2 and design.ref_audio_path and Path(design.ref_audio_path).exists():
from core.tts_service import IndexTTS2Backend
indextts2 = IndexTTS2Backend()
audio_bytes = await indextts2.generate(
text=seg.text,
spk_audio_prompt=design.ref_audio_path,
output_path=str(audio_path),
emo_text=seg.emo_text or None,
emo_alpha=seg.emo_alpha if seg.emo_text else 0.5,
)
else:
if design.voice_cache_id:
from core.cache_manager import VoiceCacheManager
@@ -688,8 +725,11 @@ async def generate_single_segment(segment_id: int, user: User, db: Session) -> N
backend_type = user.user_preferences.get("default_backend", "aliyun") if user.user_preferences else "aliyun"
user_api_key = None
if backend_type == "aliyun" and user.aliyun_api_key:
user_api_key = decrypt_api_key(user.aliyun_api_key)
if backend_type == "aliyun":
from db.crud import get_system_setting
encrypted = get_system_setting(db, "aliyun_api_key")
if encrypted:
user_api_key = decrypt_api_key(encrypted)
backend = await TTSServiceFactory.get_backend(backend_type, user_api_key)
@@ -709,7 +749,23 @@ async def generate_single_segment(segment_id: int, user: User, db: Session) -> N
audio_filename = f"ch{seg.chapter_index:03d}_seg{seg.segment_index:04d}.wav"
audio_path = output_base / audio_filename
if backend_type == "aliyun":
ref_audio_for_emo = design.ref_audio_path
if not ref_audio_for_emo:
preview_path = Path(settings.OUTPUT_DIR) / "audiobook" / str(seg.project_id) / "previews" / f"char_{char.id}.wav"
if preview_path.exists():
ref_audio_for_emo = str(preview_path)
if seg.emo_text and ref_audio_for_emo and Path(ref_audio_for_emo).exists():
from core.tts_service import IndexTTS2Backend
indextts2 = IndexTTS2Backend()
audio_bytes = await indextts2.generate(
text=seg.text,
spk_audio_prompt=ref_audio_for_emo,
output_path=str(audio_path),
emo_text=seg.emo_text,
emo_alpha=seg.emo_alpha if seg.emo_alpha is not None else 0.6,
)
elif backend_type == "aliyun":
if design.aliyun_voice_id:
audio_bytes, _ = await backend.generate_voice_design(
{"text": seg.text, "language": "zh"},
@@ -721,16 +777,6 @@ async def generate_single_segment(segment_id: int, user: User, db: Session) -> N
"language": "zh",
"instruct": _get_gendered_instruct(char.gender, design.instruct),
})
elif char.use_indextts2 and design.ref_audio_path and Path(design.ref_audio_path).exists():
from core.tts_service import IndexTTS2Backend
indextts2 = IndexTTS2Backend()
audio_bytes = await indextts2.generate(
text=seg.text,
spk_audio_prompt=design.ref_audio_path,
output_path=str(audio_path),
emo_text=seg.emo_text or None,
emo_alpha=seg.emo_alpha if seg.emo_text else 0.5,
)
else:
if design.voice_cache_id:
from core.cache_manager import VoiceCacheManager
@@ -1070,7 +1116,9 @@ async def generate_character_preview(project_id: int, char_id: int, user: User,
with open(audio_path, "wb") as f:
f.write(audio_bytes)
design.ref_audio_path = str(audio_path)
db.commit()
logger.info(f"Preview generated for char {char_id}: {audio_path}")
except Exception as e:
logger.error(f"Failed to generate preview for char {char_id}: {e}")

42
qwen3-tts-backend/core/llm_service.py
View File

@@ -1,7 +1,7 @@
import asyncio
import json
import logging
from typing import Any, Dict
from typing import Any, Callable, Dict, Optional
import httpx
@@ -14,7 +14,7 @@ class LLMService:
self.api_key = api_key
self.model = model
async def stream_chat(self, system_prompt: str, user_message: str, on_token=None, max_tokens: int = 8192) -> str:
async def stream_chat(self, system_prompt: str, user_message: str, on_token=None, max_tokens: int = 8192, usage_callback: Optional[Callable[[int, int], None]] = None) -> str:
url = f"{self.base_url}/chat/completions"
headers = {
"Authorization": f"Bearer {self.api_key}",
@@ -29,8 +29,10 @@ class LLMService:
"temperature": 0.3,
"max_tokens": max_tokens,
"stream": True,
"stream_options": {"include_usage": True},
}
full_text = ""
_usage = None
timeout = httpx.Timeout(connect=10.0, read=90.0, write=10.0, pool=5.0)
async with httpx.AsyncClient(timeout=timeout) as client:
async with client.stream("POST", url, json=payload, headers=headers) as resp:
@@ -46,6 +48,9 @@ class LLMService:
break
try:
chunk = json.loads(data)
if chunk.get("usage"):
_usage = chunk["usage"]
continue
delta = chunk["choices"][0]["delta"].get("content", "")
if delta:
full_text += delta
@@ -53,10 +58,12 @@ class LLMService:
on_token(delta)
except (json.JSONDecodeError, KeyError, IndexError):
continue
if _usage and usage_callback:
usage_callback(_usage.get("prompt_tokens", 0), _usage.get("completion_tokens", 0))
return full_text
async def stream_chat_json(self, system_prompt: str, user_message: str, on_token=None, max_tokens: int = 8192):
raw = await self.stream_chat(system_prompt, user_message, on_token, max_tokens=max_tokens)
async def stream_chat_json(self, system_prompt: str, user_message: str, on_token=None, max_tokens: int = 8192, usage_callback: Optional[Callable[[int, int], None]] = None):
raw = await self.stream_chat(system_prompt, user_message, on_token, max_tokens=max_tokens, usage_callback=usage_callback)
raw = raw.strip()
if not raw:
raise ValueError("LLM returned empty response")
@@ -74,7 +81,7 @@ class LLMService:
logger.error(f"JSON parse failed. Raw (first 500): {raw[:500]}")
raise
async def chat(self, system_prompt: str, user_message: str) -> str:
async def chat(self, system_prompt: str, user_message: str, usage_callback: Optional[Callable[[int, int], None]] = None) -> str:
url = f"{self.base_url}/chat/completions"
headers = {
"Authorization": f"Bearer {self.api_key}",
@@ -97,10 +104,13 @@ class LLMService:
logger.error(f"LLM API error {resp.status_code}: {resp.text}")
resp.raise_for_status()
data = resp.json()
usage = data.get("usage", {})
if usage and usage_callback:
usage_callback(usage.get("prompt_tokens", 0), usage.get("completion_tokens", 0))
return data["choices"][0]["message"]["content"]
async def chat_json(self, system_prompt: str, user_message: str) -> Any:
raw = await self.chat(system_prompt, user_message)
async def chat_json(self, system_prompt: str, user_message: str, usage_callback: Optional[Callable[[int, int], None]] = None) -> Any:
raw = await self.chat(system_prompt, user_message, usage_callback=usage_callback)
raw = raw.strip()
if not raw:
raise ValueError("LLM returned empty response")
@@ -118,7 +128,7 @@ class LLMService:
logger.error(f"JSON parse failed. Raw response (first 500 chars): {raw[:500]}")
raise
async def extract_characters(self, text_samples: list[str], on_token=None, on_sample=None, turbo: bool = False) -> list[Dict]:
async def extract_characters(self, text_samples: list[str], on_token=None, on_sample=None, turbo: bool = False, usage_callback: Optional[Callable[[int, int], None]] = None) -> list[Dict]:
system_prompt = (
"你是一个专业的小说分析助手兼声音导演。请分析给定的小说文本，提取所有出现的角色（包括旁白narrator）。\n"
"gender字段必须明确标注性别，只能取以下三个值之一：\"男\"、\"女\"、\"未知\"。\n"
@@ -145,7 +155,7 @@ class LLMService:
async def _extract_one(i: int, sample: str) -> list[Dict]:
user_message = f"请分析以下小说文本并提取角色：\n\n{sample}"
result = await self.stream_chat_json(system_prompt, user_message, None)
result = await self.stream_chat_json(system_prompt, user_message, None, usage_callback=usage_callback)
if on_sample:
on_sample(i, len(text_samples))
return result.get("characters", [])
@@ -160,14 +170,14 @@ class LLMService:
logger.warning(f"Character extraction failed for sample {i+1}: {r}")
else:
raw_all.extend(r)
return await self.merge_characters(raw_all)
return await self.merge_characters(raw_all, usage_callback=usage_callback)
raw_all: list[Dict] = []
for i, sample in enumerate(text_samples):
logger.info(f"Extracting characters from sample {i+1}/{len(text_samples)}")
user_message = f"请分析以下小说文本并提取角色：\n\n{sample}"
try:
result = await self.stream_chat_json(system_prompt, user_message, on_token)
result = await self.stream_chat_json(system_prompt, user_message, on_token, usage_callback=usage_callback)
raw_all.extend(result.get("characters", []))
except Exception as e:
logger.warning(f"Character extraction failed for sample {i+1}: {e}")
@@ -175,9 +185,9 @@ class LLMService:
on_sample(i, len(text_samples))
if len(text_samples) == 1:
return raw_all
return await self.merge_characters(raw_all)
return await self.merge_characters(raw_all, usage_callback=usage_callback)
async def merge_characters(self, raw_characters: list[Dict]) -> list[Dict]:
async def merge_characters(self, raw_characters: list[Dict], usage_callback: Optional[Callable[[int, int], None]] = None) -> list[Dict]:
system_prompt = (
"你是一个专业的小说角色整合助手。你收到的是从同一本书不同段落中提取的角色列表，其中可能存在重复。\n"
"请完成以下任务：\n"
@@ -191,7 +201,7 @@ class LLMService:
)
user_message = f"请整合以下角色列表：\n\n{json.dumps(raw_characters, ensure_ascii=False, indent=2)}"
try:
result = await self.chat_json(system_prompt, user_message)
result = await self.chat_json(system_prompt, user_message, usage_callback=usage_callback)
return result.get("characters", [])
except Exception as e:
logger.warning(f"Character merge failed, falling back to name-dedup: {e}")
@@ -202,7 +212,7 @@ class LLMService:
seen[name] = c
return list(seen.values())
async def parse_chapter_segments(self, chapter_text: str, character_names: list[str], on_token=None) -> list[Dict]:
async def parse_chapter_segments(self, chapter_text: str, character_names: list[str], on_token=None, usage_callback: Optional[Callable[[int, int], None]] = None) -> list[Dict]:
names_str = "、".join(character_names)
system_prompt = (
"你是一个专业的有声书制作助手。请将给定的章节文本解析为对话片段列表。"
@@ -217,7 +227,7 @@ class LLMService:
'{"character": "角色名", "text": "对话内容", "emo_text": "开心", "emo_alpha": 0.6}, ...]'
)
user_message = f"请解析以下章节文本：\n\n{chapter_text}"
result = await self.stream_chat_json(system_prompt, user_message, on_token, max_tokens=16384)
result = await self.stream_chat_json(system_prompt, user_message, on_token, max_tokens=16384, usage_callback=usage_callback)
if isinstance(result, list):
return result
return []