feat: align surgery API with schemas and extend client tooling
- Refactor app API and schemas; adjust surgery pipeline, repository, and session manager. - Improve consumption TSV logging and consumable vision integration; trim voice resolution. - Add Baidu Face 1:N search script, .env.example entries, and client API integration doc. - Update demo client, staging checklist, surgery interface doc, and related tests; add sample face image. Made-with: Cursor
This commit is contained in:
Kevin
201
app/api.py
201
app/api.py
@@ -2,8 +2,7 @@ import asyncio
|
||||
from collections.abc import Awaitable, Callable
|
||||
from typing import Annotated
|
||||
|
||||
from fastapi import APIRouter, Depends, File, HTTPException, Path, Query, UploadFile, status
|
||||
from fastapi.responses import Response
|
||||
from fastapi import APIRouter, Depends, File, HTTPException, Path, UploadFile, status
|
||||
from fastapi.responses import JSONResponse
|
||||
from loguru import logger
|
||||
from sqlalchemy.exc import SQLAlchemyError
|
||||
@@ -18,12 +17,8 @@ from app.schemas import (
|
||||
SurgeryEndRequest,
|
||||
SurgeryPendingConfirmationResolveResponse,
|
||||
SurgeryPendingConfirmationResponse,
|
||||
SurgeryPendingResolveTextRequest,
|
||||
SurgeryResultResponse,
|
||||
SurgeryStartRequest,
|
||||
SurgeryVoiceAuditItem,
|
||||
SurgeryVoiceAuditsListResponse,
|
||||
SurgeryVoiceStatusResponse,
|
||||
build_consumption_summary,
|
||||
)
|
||||
from app.services.surgery_pipeline import SurgeryPipeline
|
||||
@@ -251,7 +246,11 @@ async def end_surgery(
|
||||
"根据手术 6 位号查询该台手术的耗材消耗明细（多行）及按物品汇总。"
|
||||
"手术进行中返回当前内存已记账结果；结束后返回数据库持久化结果。"
|
||||
"若手术从未开始或尚无可查的最终归档，返回 503。"
|
||||
"使用 GET：只读、幂等。"
|
||||
"使用 GET：只读、幂等。\n\n"
|
||||
"响应体 `details` 与 `summary` 的字段定义见模式 SurgeryConsumptionDetail / SurgeryConsumptionSummary；"
|
||||
"若服务端启用耗材 TSV 文本日志，文件明细列为 tab 分隔的 "
|
||||
"item_id、item_name、qty、doctor_id、timestamp（文末另有仅三列的汇总块 item_id、item_name、qty），"
|
||||
"与 HTTP JSON 字段一致。"
|
||||
),
|
||||
)
|
||||
async def get_surgery_result(
|
||||
@@ -296,13 +295,22 @@ async def get_surgery_result(
|
||||
"description": "当前无待确认项或手术未在进行。",
|
||||
"model": SurgeryClientErrorResponse,
|
||||
},
|
||||
status.HTTP_422_UNPROCESSABLE_CONTENT: {
|
||||
"description": "提示文本为空等导致无法合成播报。",
|
||||
"model": SurgeryClientErrorResponse,
|
||||
},
|
||||
status.HTTP_503_SERVICE_UNAVAILABLE: {
|
||||
"description": "百度语音未配置或 TTS 调用失败。",
|
||||
"model": SurgeryClientErrorResponse,
|
||||
},
|
||||
},
|
||||
tags=["client"],
|
||||
summary="拉取待确认耗材",
|
||||
summary="拉取待确认耗材（含 TTS 音频）",
|
||||
description=(
|
||||
"返回当前 FIFO 队首的一条低置信度识别。"
|
||||
"客户端应播报 prompt_text 并由医生确认后调用 resolve 接口。"
|
||||
"无待确认项时返回 404。"
|
||||
"返回当前 FIFO 队首的一条低置信度识别；"
|
||||
"响应内 `prompt_audio_mp3_base64` 为与 `prompt_text` 一致的 MP3（Base64），客户端可直接解码播放。"
|
||||
"无待确认项时返回 404；合成失败或未配置语音服务时返回 422/503（见错误码）。"
|
||||
"医生确认后请使用 `POST .../resolve` 上传 WAV。"
|
||||
),
|
||||
)
|
||||
async def get_pending_consumable_confirmation(
|
||||
@@ -317,7 +325,10 @@ async def get_pending_consumable_confirmation(
|
||||
],
|
||||
pipeline: Annotated[SurgeryPipeline, Depends(get_surgery_pipeline)],
|
||||
) -> SurgeryPendingConfirmationResponse:
|
||||
payload = pipeline.get_pending_confirmation_for_client(surgery_id)
|
||||
try:
|
||||
payload = await pipeline.get_pending_confirmation_for_client(surgery_id)
|
||||
except SurgeryPipelineError as exc:
|
||||
_raise_confirmation_http(exc, surgery_id)
|
||||
if payload is None:
|
||||
raise HTTPException(
|
||||
status_code=status.HTTP_404_NOT_FOUND,
|
||||
@@ -344,7 +355,7 @@ async def get_pending_consumable_confirmation(
|
||||
description=(
|
||||
"multipart/form-data 上传单个 WAV 文件（字段名 `audio`）。"
|
||||
"服务端将音频存入 MinIO、调用百度 ASR 识别、解析候选项并完成确认。"
|
||||
"记一条 source=voice 的消耗；若语音表示否认全部候选则不记消耗。"
|
||||
"解析并确认后记一条消耗明细；若语音表示否认全部候选则不记消耗。"
|
||||
),
|
||||
)
|
||||
async def resolve_pending_consumable_confirmation(
|
||||
@@ -407,167 +418,3 @@ async def resolve_pending_consumable_confirmation(
|
||||
asr_text=result.asr_text,
|
||||
audio_object_key=result.audio_object_key,
|
||||
)
|
||||
|
||||
|
||||
@router.post(
|
||||
"/client/surgeries/{surgery_id}/pending-confirmation/{confirmation_id}/resolve-text",
|
||||
response_model=SurgeryPendingConfirmationResolveResponse,
|
||||
responses={
|
||||
status.HTTP_404_NOT_FOUND: {"model": SurgeryClientErrorResponse},
|
||||
status.HTTP_409_CONFLICT: {"model": SurgeryClientErrorResponse},
|
||||
status.HTTP_422_UNPROCESSABLE_CONTENT: {"model": SurgeryClientErrorResponse},
|
||||
},
|
||||
tags=["client"],
|
||||
summary="提交客户端语音识别文本以确认耗材",
|
||||
description=(
|
||||
"由浏览器 Web Speech 等本机 STT 得到的文本，不做 MinIO/百度 ASR；"
|
||||
"候选项解析与上传 WAV 接口一致。"
|
||||
),
|
||||
)
|
||||
async def resolve_pending_consumable_confirmation_text(
|
||||
surgery_id: Annotated[
|
||||
str,
|
||||
Path(
|
||||
min_length=6,
|
||||
max_length=6,
|
||||
pattern=r"^\d{6}$",
|
||||
description="手术 6 位号，仅允许 6 位数字。",
|
||||
),
|
||||
],
|
||||
confirmation_id: Annotated[str, Path(min_length=1, max_length=128)],
|
||||
body: SurgeryPendingResolveTextRequest,
|
||||
pipeline: Annotated[SurgeryPipeline, Depends(get_surgery_pipeline)],
|
||||
) -> SurgeryPendingConfirmationResolveResponse:
|
||||
try:
|
||||
result = await pipeline.resolve_pending_confirmation_from_client_text(
|
||||
surgery_id=surgery_id,
|
||||
confirmation_id=confirmation_id,
|
||||
recognized_text=body.recognized_text,
|
||||
)
|
||||
except SurgeryPipelineError as exc:
|
||||
_raise_confirmation_http(exc, surgery_id)
|
||||
return SurgeryPendingConfirmationResolveResponse(
|
||||
surgery_id=surgery_id,
|
||||
confirmation_id=confirmation_id,
|
||||
status="accepted",
|
||||
message=result.message,
|
||||
resolved_label=result.resolved_label,
|
||||
rejected=result.rejected,
|
||||
asr_text=result.asr_text,
|
||||
audio_object_key=result.audio_object_key,
|
||||
)
|
||||
|
||||
|
||||
@router.get(
|
||||
"/client/surgeries/{surgery_id}/pending-confirmation/{confirmation_id}/prompt-audio",
|
||||
responses={
|
||||
status.HTTP_404_NOT_FOUND: {"model": SurgeryClientErrorResponse},
|
||||
status.HTTP_422_UNPROCESSABLE_CONTENT: {"model": SurgeryClientErrorResponse},
|
||||
status.HTTP_503_SERVICE_UNAVAILABLE: {"model": SurgeryClientErrorResponse},
|
||||
},
|
||||
tags=["client"],
|
||||
summary="待确认话术的 TTS 音频（MP3）",
|
||||
description="使用百度在线合成，与 prompt_text 一致；供浏览器 MediaElement 直放。未配置百度语音时返回 503。",
|
||||
response_class=Response,
|
||||
)
|
||||
async def get_pending_prompt_audio_mpeg(
|
||||
surgery_id: Annotated[
|
||||
str,
|
||||
Path(
|
||||
min_length=6,
|
||||
max_length=6,
|
||||
pattern=r"^\d{6}$",
|
||||
description="手术 6 位号，仅允许 6 位数字。",
|
||||
),
|
||||
],
|
||||
confirmation_id: Annotated[str, Path(min_length=1, max_length=128)],
|
||||
pipeline: Annotated[SurgeryPipeline, Depends(get_surgery_pipeline)],
|
||||
) -> Response:
|
||||
try:
|
||||
data = await pipeline.get_pending_prompt_audio_mp3(
|
||||
surgery_id=surgery_id,
|
||||
confirmation_id=confirmation_id,
|
||||
)
|
||||
except SurgeryPipelineError as exc:
|
||||
_raise_confirmation_http(exc, surgery_id)
|
||||
return Response(
|
||||
content=data,
|
||||
media_type="audio/mpeg",
|
||||
headers={"Cache-Control": "no-store"},
|
||||
)
|
||||
|
||||
|
||||
@router.get(
|
||||
"/internal/surgeries/{surgery_id}/voice-status",
|
||||
response_model=SurgeryVoiceStatusResponse,
|
||||
tags=["internal"],
|
||||
summary="人工确认队列状态（联调）",
|
||||
description="查询指定进行中手术的待确认队列长度与最近话术摘要。手术未在进行返回 404。",
|
||||
)
|
||||
async def get_surgery_voice_status(
|
||||
surgery_id: Annotated[
|
||||
str,
|
||||
Path(
|
||||
min_length=6,
|
||||
max_length=6,
|
||||
pattern=r"^\d{6}$",
|
||||
description="手术 6 位号，仅允许 6 位数字。",
|
||||
),
|
||||
],
|
||||
pipeline: Annotated[SurgeryPipeline, Depends(get_surgery_pipeline)],
|
||||
) -> SurgeryVoiceStatusResponse:
|
||||
payload = pipeline.voice_status(surgery_id)
|
||||
if payload is None:
|
||||
raise HTTPException(
|
||||
status_code=status.HTTP_404_NOT_FOUND,
|
||||
detail={
|
||||
"code": "SURGERY_NOT_ACTIVE",
|
||||
"message": "该手术当前不在进行中，无实时语音状态。",
|
||||
"surgery_id": surgery_id,
|
||||
},
|
||||
)
|
||||
return SurgeryVoiceStatusResponse(
|
||||
surgery_id=surgery_id,
|
||||
voice_enabled=bool(payload["voice_enabled"]),
|
||||
pending_queue_approx=int(payload["pending_queue_approx"]),
|
||||
last_prompt_snippet=payload.get("last_prompt_snippet"),
|
||||
last_asr_text=payload.get("last_asr_text"),
|
||||
last_error=payload.get("last_error"),
|
||||
)
|
||||
|
||||
|
||||
@router.get(
|
||||
"/internal/surgeries/{surgery_id}/voice-audits",
|
||||
response_model=SurgeryVoiceAuditsListResponse,
|
||||
tags=["internal"],
|
||||
summary="语音确认审计记录（按手术号分页）",
|
||||
description=(
|
||||
"查询持久化表 `voice_confirmation_audits`：ASR 文本、解析结果、"
|
||||
"候选项快照、MinIO 对象键、失败原因等。用于追溯、对账与报表；"
|
||||
"不区分手术是否仍进行中，只要库里有记录即返回。"
|
||||
),
|
||||
)
|
||||
async def get_surgery_voice_audits(
|
||||
surgery_id: Annotated[
|
||||
str,
|
||||
Path(
|
||||
min_length=6,
|
||||
max_length=6,
|
||||
pattern=r"^\d{6}$",
|
||||
description="手术 6 位号，仅允许 6 位数字。",
|
||||
),
|
||||
],
|
||||
pipeline: Annotated[SurgeryPipeline, Depends(get_surgery_pipeline)],
|
||||
limit: Annotated[int, Query(ge=1, le=200, description="每页条数。")] = 50,
|
||||
offset: Annotated[int, Query(ge=0, description="跳过前若干条，供分页。")] = 0,
|
||||
) -> SurgeryVoiceAuditsListResponse:
|
||||
rows, total = await pipeline.list_voice_audits(
|
||||
surgery_id, limit=limit, offset=offset
|
||||
)
|
||||
return SurgeryVoiceAuditsListResponse(
|
||||
surgery_id=surgery_id,
|
||||
total=total,
|
||||
limit=limit,
|
||||
offset=offset,
|
||||
items=[SurgeryVoiceAuditItem.model_validate(r) for r in rows],
|
||||
)
|
||||
|
||||
@@ -101,7 +101,7 @@ class Settings(BaseSettings):
|
||||
video_result_doctor_id: str = "vision"
|
||||
#: 为 true 时，每次单帧分类得到 top1 等结果会打一条 INFO 日志（联调用；高流量时建议关）。
|
||||
video_log_inference_results: bool = False
|
||||
#: 为 true 时，将时间窗级识别写入文本日志（`start_surgery` 时按手术截断/初始化，窗内结果追加；Top2/3 仅名称；数量恒 1）。
|
||||
#: 为 true 时，将时间窗级识别写入文本日志（`start_surgery` 时按手术截断/初始化；每行 tab：item_id、item_name、qty、doctor_id、timestamp；停录后追加汇总块 item_id、item_name、qty）。
|
||||
consumption_tsv_log_enabled: bool = True
|
||||
#: 路径模板，须含 `{surgery_id}`（每例手术独立文件）。不含占位时自动在扩展名前追加 `_<surgery_id>`。
|
||||
consumption_tsv_log_path: str = "logs/consumption_{surgery_id}.txt"
|
||||
|
||||
@@ -6,7 +6,7 @@ from sqlalchemy import delete, select
|
||||
from sqlalchemy.ext.asyncio import AsyncSession
|
||||
|
||||
from app.db.models import SurgeryFinalResult, SurgeryResultDetailRow
|
||||
from app.schemas import SurgeryConsumptionDetail
|
||||
from app.schemas import SurgeryConsumptionDetail, SurgeryConsumptionStored
|
||||
|
||||
|
||||
class SurgeryResultRepository:
|
||||
@@ -17,7 +17,7 @@ class SurgeryResultRepository:
|
||||
session: AsyncSession,
|
||||
*,
|
||||
surgery_id: str,
|
||||
details: list[SurgeryConsumptionDetail],
|
||||
details: list[SurgeryConsumptionStored],
|
||||
completed_at: datetime | None = None,
|
||||
) -> None:
|
||||
when = completed_at or datetime.now(timezone.utc)
|
||||
@@ -37,7 +37,7 @@ class SurgeryResultRepository:
|
||||
surgery_id=surgery_id,
|
||||
item_id=d.item_id,
|
||||
item_name=d.item_name,
|
||||
quantity=d.quantity,
|
||||
quantity=d.qty,
|
||||
doctor_id=d.doctor_id,
|
||||
recorded_at=d.timestamp,
|
||||
source=d.source,
|
||||
@@ -64,10 +64,9 @@ class SurgeryResultRepository:
|
||||
SurgeryConsumptionDetail(
|
||||
item_id=r.item_id,
|
||||
item_name=r.item_name,
|
||||
quantity=r.quantity,
|
||||
qty=r.quantity,
|
||||
doctor_id=r.doctor_id,
|
||||
timestamp=r.recorded_at,
|
||||
source=r.source,
|
||||
)
|
||||
for r in rows
|
||||
]
|
||||
|
||||
176
app/schemas.py
176
app/schemas.py
@@ -1,5 +1,6 @@
|
||||
from __future__ import annotations
|
||||
|
||||
from dataclasses import dataclass
|
||||
from datetime import datetime
|
||||
|
||||
from pydantic import BaseModel, ConfigDict, Field
|
||||
@@ -75,24 +76,72 @@ class SurgeryClientErrorResponse(BaseModel):
|
||||
|
||||
|
||||
class SurgeryConsumptionDetail(BaseModel):
|
||||
"""单条消耗明细（按事件发生，可能多行）。"""
|
||||
"""单条消耗明细（HTTP 与 OpenAPI；按事件发生，可能多行）。
|
||||
|
||||
item_id: str = Field(description="物品 ID。")
|
||||
item_name: str = Field(description="物品名称。")
|
||||
quantity: int = Field(ge=0, description="本条记录对应的消耗数量。")
|
||||
doctor_id: str = Field(description="医生 ID。")
|
||||
timestamp: datetime = Field(description="记录时间（ISO 8601）。")
|
||||
source: str = Field(
|
||||
default="vision",
|
||||
description="记录来源：vision 自动识别；voice 语音确认。",
|
||||
JSON 字段顺序：item_id → item_name → qty → doctor_id → timestamp。
|
||||
（可选落盘耗材 TSV 明细列与此一致：item_id、item_name、qty、doctor_id、timestamp。）
|
||||
"""
|
||||
|
||||
model_config = ConfigDict(
|
||||
json_schema_extra={
|
||||
"example": {
|
||||
"item_id": "HC001",
|
||||
"item_name": "纱布",
|
||||
"qty": 1,
|
||||
"doctor_id": "D1001",
|
||||
"timestamp": "2026-04-21T10:30:00+08:00",
|
||||
}
|
||||
}
|
||||
)
|
||||
|
||||
item_id: str = Field(
|
||||
description=(
|
||||
"业务物品标识：优先为耗材目录中的产品编码；"
|
||||
"目录键经名称归一化后与分类类名匹配，未命中目录时与模型输出类名一致。"
|
||||
),
|
||||
)
|
||||
item_name: str = Field(description="物品名称（分类或确认后的展示名）。")
|
||||
qty: int = Field(ge=0, description="本条记录对应的消耗数量。")
|
||||
doctor_id: str = Field(description="医生 ID。")
|
||||
timestamp: datetime = Field(description="记录时间（ISO 8601，date-time）。")
|
||||
|
||||
|
||||
@dataclass
|
||||
class SurgeryConsumptionStored:
|
||||
"""内存 / 数据库持久化用的明细行（含 source，仅服务端内部使用，不随 HTTP 返回）。"""
|
||||
|
||||
item_id: str
|
||||
item_name: str
|
||||
qty: int
|
||||
doctor_id: str
|
||||
timestamp: datetime
|
||||
source: str = "vision"
|
||||
|
||||
def as_response(self) -> SurgeryConsumptionDetail:
|
||||
return SurgeryConsumptionDetail(
|
||||
item_id=self.item_id,
|
||||
item_name=self.item_name,
|
||||
qty=self.qty,
|
||||
doctor_id=self.doctor_id,
|
||||
timestamp=self.timestamp,
|
||||
)
|
||||
|
||||
|
||||
class SurgeryConsumptionSummary(BaseModel):
|
||||
"""按物品汇总：该手术下该物品消耗数量合计。"""
|
||||
"""按物品汇总：该手术下该物品消耗数量合计（item_id、item_name、total_quantity）。"""
|
||||
|
||||
item_id: str = Field(description="物品 ID。")
|
||||
item_name: str = Field(description="物品名称。")
|
||||
model_config = ConfigDict(
|
||||
json_schema_extra={
|
||||
"example": {
|
||||
"item_id": "HC001",
|
||||
"item_name": "纱布",
|
||||
"total_quantity": 3,
|
||||
}
|
||||
}
|
||||
)
|
||||
|
||||
item_id: str = Field(description="物品 ID，与明细中 item_id 一致。")
|
||||
item_name: str = Field(description="物品名称，取该物品首条明细中的名称。")
|
||||
total_quantity: int = Field(ge=0, description="该物品在本台手术中的消耗数量合计。")
|
||||
|
||||
|
||||
@@ -105,7 +154,7 @@ def build_consumption_summary(
|
||||
if row.item_id not in totals:
|
||||
totals[row.item_id] = (row.item_name, 0)
|
||||
name, acc = totals[row.item_id]
|
||||
totals[row.item_id] = (name, acc + row.quantity)
|
||||
totals[row.item_id] = (name, acc + row.qty)
|
||||
return [
|
||||
SurgeryConsumptionSummary(
|
||||
item_id=iid,
|
||||
@@ -116,73 +165,6 @@ def build_consumption_summary(
|
||||
]
|
||||
|
||||
|
||||
class SurgeryVoiceStatusResponse(BaseModel):
|
||||
"""手术进行中人工确认（客户端播报）联调状态。"""
|
||||
|
||||
surgery_id: str = Field(description="手术 6 位号。")
|
||||
voice_enabled: bool = Field(
|
||||
description="是否启用了低置信度人工确认（客户端拉取待确认项）。",
|
||||
)
|
||||
pending_queue_approx: int = Field(
|
||||
ge=0,
|
||||
description="待医生确认的追问任务数量（FIFO 队列长度）。",
|
||||
)
|
||||
last_prompt_snippet: str | None = Field(
|
||||
default=None,
|
||||
description="最近一次生成的待确认话术摘要。",
|
||||
)
|
||||
last_asr_text: str | None = Field(
|
||||
default=None,
|
||||
description="最近一次语音确认接口产生的 ASR 文本。",
|
||||
)
|
||||
last_error: str | None = Field(
|
||||
default=None,
|
||||
description="最近一次语音确认错误说明（如 ASR/解析失败）。",
|
||||
)
|
||||
|
||||
|
||||
class SurgeryVoiceAuditItem(BaseModel):
|
||||
"""单条 `voice_confirmation_audits` 行（追溯对账用）。"""
|
||||
|
||||
id: int
|
||||
confirmation_id: str
|
||||
status: str = Field(
|
||||
description=(
|
||||
"recognized / rejected / parse_failed / asr_failed / invalid_audio / "
|
||||
"upload_failed / client_stt_empty / client_stt_parse_failed 等"
|
||||
),
|
||||
)
|
||||
audio_object_key: str | None = None
|
||||
audio_content_type: str | None = None
|
||||
audio_size_bytes: int | None = None
|
||||
audio_sha256: str | None = None
|
||||
asr_text: str | None = None
|
||||
resolved_label: str | None = None
|
||||
options_snapshot_json: str | None = Field(
|
||||
default=None,
|
||||
description="当次候选项与置信度 JSON 快照。",
|
||||
)
|
||||
error_message: str | None = None
|
||||
created_at: datetime = Field(
|
||||
description="记录写入时间（UTC）。",
|
||||
)
|
||||
|
||||
model_config = ConfigDict(from_attributes=True)
|
||||
|
||||
|
||||
class SurgeryVoiceAuditsListResponse(BaseModel):
|
||||
"""按手术号分页的语音确认审计列表。"""
|
||||
|
||||
surgery_id: str
|
||||
total: int = Field(ge=0, description="该手术在表中的总条数（不受本页 limit 截断）。")
|
||||
limit: int = Field(ge=1, le=200)
|
||||
offset: int = Field(ge=0)
|
||||
items: list[SurgeryVoiceAuditItem] = Field(
|
||||
default_factory=list,
|
||||
description="按 `created_at` 降序。",
|
||||
)
|
||||
|
||||
|
||||
class PendingConfirmationOption(BaseModel):
|
||||
label: str
|
||||
confidence: float
|
||||
@@ -193,23 +175,19 @@ class SurgeryPendingConfirmationResponse(BaseModel):
|
||||
|
||||
surgery_id: str
|
||||
confirmation_id: str
|
||||
prompt_text: str = Field(description="可直接用于 TTS 播报的话术。")
|
||||
prompt_text: str = Field(description="可直接用于展示或无障碍朗读的话术（与 MP3 内容一致）。")
|
||||
prompt_audio_mp3_base64: str = Field(
|
||||
description=(
|
||||
"与 prompt_text 对应的百度 TTS 音频（MP3）的标准 Base64 字符串（无换行）；"
|
||||
"客户端解码为二进制后以 audio/mpeg 播放。"
|
||||
),
|
||||
)
|
||||
options: list[PendingConfirmationOption]
|
||||
model_top1_label: str = Field(description="模型原始 Top1 标签（可能不在候选清单内）。")
|
||||
model_top1_confidence: float
|
||||
created_at: datetime
|
||||
|
||||
|
||||
class SurgeryPendingResolveTextRequest(BaseModel):
|
||||
"""由浏览器 Web Speech 等客户端本地识别后提交的文本，语义与经百度 ASR 得到的文本相同。"""
|
||||
|
||||
recognized_text: str = Field(
|
||||
min_length=1,
|
||||
max_length=2000,
|
||||
description="识别文本；服务端用与语音接口相同的规则解析候选项。",
|
||||
)
|
||||
|
||||
|
||||
class SurgeryPendingConfirmationResolveResponse(BaseModel):
|
||||
surgery_id: str
|
||||
confirmation_id: str
|
||||
@@ -244,21 +222,21 @@ class SurgeryResultResponse(BaseModel):
|
||||
{
|
||||
"item_id": "HC001",
|
||||
"item_name": "纱布",
|
||||
"quantity": 2,
|
||||
"qty": 2,
|
||||
"doctor_id": "D1001",
|
||||
"timestamp": "2026-04-21T10:30:00+08:00",
|
||||
},
|
||||
{
|
||||
"item_id": "HC001",
|
||||
"item_name": "纱布",
|
||||
"quantity": 1,
|
||||
"qty": 1,
|
||||
"doctor_id": "D1002",
|
||||
"timestamp": "2026-04-21T11:05:00+08:00",
|
||||
},
|
||||
{
|
||||
"item_id": "HC002",
|
||||
"item_name": "缝线",
|
||||
"quantity": 1,
|
||||
"qty": 1,
|
||||
"doctor_id": "D1001",
|
||||
"timestamp": "2026-04-21T10:45:00+08:00",
|
||||
},
|
||||
@@ -276,9 +254,15 @@ class SurgeryResultResponse(BaseModel):
|
||||
message: str = Field(description="返回说明。")
|
||||
details: list[SurgeryConsumptionDetail] = Field(
|
||||
default_factory=list,
|
||||
description="消耗明细行：每条含物品、数量、医生与时间；同一物品可多次出现。",
|
||||
description=(
|
||||
"消耗明细（多行）。每行字段顺序：item_id、item_name、qty、doctor_id、timestamp；"
|
||||
"同一 item_id 可多次出现。"
|
||||
),
|
||||
)
|
||||
summary: list[SurgeryConsumptionSummary] = Field(
|
||||
default_factory=list,
|
||||
description="按物品汇总的消耗合计，应与 details 按 item_id 汇总一致。",
|
||||
description=(
|
||||
"按 item_id 汇总的合计表（仅 item_id、item_name、total_quantity），"
|
||||
"应与 details 按 item_id 汇总 qty 一致。"
|
||||
),
|
||||
)
|
||||
|
||||
@@ -219,6 +219,13 @@ def cls_top3_from_result(
|
||||
n3 = str(cls.names.get(int(t5i[2]), "")).strip()
|
||||
c3 = _ci(2)
|
||||
|
||||
def _pid(label: str) -> str:
|
||||
lb = (label or "").strip()
|
||||
if not lb:
|
||||
return ""
|
||||
norm = _norm_product_name(lb)
|
||||
return (name_to_code.get(norm) or name_to_code.get(lb) or "").strip()
|
||||
|
||||
return ClsTop3(
|
||||
t1_name=n1,
|
||||
t1_conf=c1,
|
||||
@@ -226,9 +233,9 @@ def cls_top3_from_result(
|
||||
t2_conf=c2,
|
||||
t3_name=n3,
|
||||
t3_conf=c3,
|
||||
t1_pid=name_to_code.get(n1, ""),
|
||||
t2_pid=name_to_code.get(n2, ""),
|
||||
t3_pid=name_to_code.get(n3, ""),
|
||||
t1_pid=_pid(n1),
|
||||
t2_pid=_pid(n2),
|
||||
t3_pid=_pid(n3),
|
||||
)
|
||||
|
||||
|
||||
|
||||
@@ -1,4 +1,6 @@
|
||||
"""每例手术一个文本文件（制表符列）：`start_surgery` 时截断并写表头，每次时间窗识别**追加**一行。终端 Markdown 中时间戳为可读形式；落盘行内仍为 ISO 便于程序解析。
|
||||
"""每例手术一个文本文件（制表符列）：`start_surgery` 时截断并写表头，每次时间窗识别**追加**一行（仅 item_id, item_name, qty, doctor_id, timestamp）。终端 Markdown 时间戳为可读形式；落盘时间戳为 ISO 区间便于程序解析。
|
||||
|
||||
手术结束时再追加一节汇总行：item_id, item_name, qty（无其它列）。
|
||||
|
||||
时间戳：在拉流起点记录 `time.time()`，与 `time.monotonic()` 时间窗对齐。直播 RTSP 经 OpenCV 一般无可靠绝对时码，以本机接收时刻为准。
|
||||
"""
|
||||
@@ -14,11 +16,12 @@ from zoneinfo import ZoneInfo, ZoneInfoNotFoundError
|
||||
from loguru import logger
|
||||
|
||||
from app.config import settings
|
||||
from app.services.consumable_vision_algorithm import ClsTop3
|
||||
from app.services.consumable_vision_algorithm import ClsTop3, _norm_product_name
|
||||
from app.terminal_markdown import print_markdown_stderr
|
||||
|
||||
# 制表符分隔；时间范围用 U+2013 连接；Top2/3 仅名称；本窗消耗数量恒为 1
|
||||
HEADER = "物品id\t物品名称\tTop2物品名称\tTop3物品名称\t消耗数量\t医生id\t时间戳\n"
|
||||
# 制表符分隔；时间范围用 U+2013 连接；本窗消耗数量恒为 1
|
||||
HEADER = "item_id\titem_name\tqty\tdoctor_id\ttimestamp\n"
|
||||
SUMMARY_HEADER = "item_id\titem_name\tqty\n"
|
||||
_RANGE_SEP = "\u2013" # en dash，与样例 `00:00:00.000–00:00:45.000` 一致
|
||||
|
||||
_lock = threading.Lock()
|
||||
@@ -83,13 +86,20 @@ def _encode_cell(value: str) -> str:
|
||||
return s
|
||||
|
||||
|
||||
def _item_id_for_row(name: str, pid: str, name_to_code: dict[str, str]) -> str:
|
||||
p = (pid or "").strip()
|
||||
def resolve_consumption_item_id(
|
||||
t1_name: str,
|
||||
t1_pid: str,
|
||||
name_to_code: dict[str, str],
|
||||
) -> str:
|
||||
"""业务物品 id：`name_to_code` 的键为归一化名称，须与分类输出一同参与查找。"""
|
||||
n = (t1_name or "").strip()
|
||||
norm = _norm_product_name(n)
|
||||
code = (name_to_code.get(norm) or name_to_code.get(n) or "").strip()
|
||||
if code:
|
||||
return code
|
||||
p = (t1_pid or "").strip()
|
||||
if p:
|
||||
return p
|
||||
n = (name or "").strip()
|
||||
if n in name_to_code:
|
||||
return (name_to_code.get(n) or n).strip()
|
||||
return n
|
||||
|
||||
|
||||
@@ -102,17 +112,12 @@ def build_tsv_line(
|
||||
wall_start_epoch: float,
|
||||
wall_end_epoch: float,
|
||||
) -> str:
|
||||
id1 = _item_id_for_row(best.t1_name, best.t1_pid, name_to_code)
|
||||
# 与历史样例：Top1 为「名称 置信度」四位小数
|
||||
name1 = f"{(best.t1_name or '').strip()} {best.t1_conf:.4f}".strip()
|
||||
n2 = (best.t2_name or "").strip()
|
||||
n3 = (best.t3_name or "").strip()
|
||||
id1 = resolve_consumption_item_id(best.t1_name, best.t1_pid, name_to_code)
|
||||
name1 = (best.t1_name or "").strip()
|
||||
ts = format_consumption_timestamp(camera_id, wall_start_epoch, wall_end_epoch)
|
||||
row = [
|
||||
_encode_cell(id1),
|
||||
_encode_cell(name1),
|
||||
_encode_cell(n2),
|
||||
_encode_cell(n3),
|
||||
"1",
|
||||
_encode_cell(doctor_id),
|
||||
_encode_cell(ts),
|
||||
@@ -179,25 +184,17 @@ def build_consumption_markdown(
|
||||
wall_start_epoch: float,
|
||||
wall_end_epoch: float,
|
||||
) -> str:
|
||||
"""终端用：Top1 含 id/名称/置信度；Top2/3 仅名称；消耗数量恒为 1。"""
|
||||
id1 = _item_id_for_row(best.t1_name, best.t1_pid, name_to_code)
|
||||
"""终端用：与落盘列一致；本窗 qty 恒为 1。"""
|
||||
id1 = resolve_consumption_item_id(best.t1_name, best.t1_pid, name_to_code)
|
||||
n1 = (best.t1_name or "").strip()
|
||||
has2 = bool((best.t2_name or "").strip())
|
||||
has3 = bool((best.t3_name or "").strip())
|
||||
n2 = (best.t2_name or "").strip() if has2 else ""
|
||||
n3 = (best.t3_name or "").strip() if has3 else ""
|
||||
dash = "—"
|
||||
ts = format_consumption_timestamp_readable(camera_id, wall_start_epoch, wall_end_epoch)
|
||||
return "\n".join(
|
||||
[
|
||||
"| Top1 物品id | Top1 物品名称 | Top1 置信度 | Top2 物品名称 | Top3 物品名称 | 消耗数量 | 医生id | 时间戳 |",
|
||||
"| :--- | :--- | ---: | :--- | :--- | ---: | :--- | :--- |",
|
||||
"| {} | {} | {:.4f} | {} | {} | 1 | {} | {} |".format(
|
||||
"| item_id | item_name | qty | doctor_id | timestamp |",
|
||||
"| :--- | :--- | ---: | :--- | :--- |",
|
||||
"| {} | {} | 1 | {} | {} |".format(
|
||||
_md_cell(id1),
|
||||
_md_cell(n1),
|
||||
best.t1_conf,
|
||||
_md_cell(n2) if has2 else dash,
|
||||
_md_cell(n3) if has3 else dash,
|
||||
_md_cell(doctor_id),
|
||||
_md_cell(ts),
|
||||
),
|
||||
@@ -206,6 +203,47 @@ def build_consumption_markdown(
|
||||
)
|
||||
|
||||
|
||||
def append_consumption_log_summary(
|
||||
surgery_id: str,
|
||||
totals: dict[str, tuple[str, int]],
|
||||
) -> None:
|
||||
"""在明细行之后追加汇总块（表头 + 每物品一行）。"""
|
||||
if not settings.consumption_tsv_log_enabled or not totals:
|
||||
return
|
||||
path = resolved_consumption_log_path(surgery_id)
|
||||
if not path.is_file():
|
||||
return
|
||||
body = "".join(
|
||||
["\n", SUMMARY_HEADER]
|
||||
+ [
|
||||
"\t".join([_encode_cell(iid), _encode_cell(name), str(qty)]) + "\n"
|
||||
for iid, (name, qty) in sorted(totals.items(), key=lambda x: x[0])
|
||||
]
|
||||
)
|
||||
with _lock:
|
||||
with path.open("a", encoding="utf-8") as f:
|
||||
f.write(body)
|
||||
|
||||
|
||||
def print_consumption_summary_markdown(
|
||||
totals: dict[str, tuple[str, int]],
|
||||
) -> None:
|
||||
if not settings.consumption_log_markdown_terminal or not totals:
|
||||
return
|
||||
lines = [
|
||||
"## 消耗汇总",
|
||||
"",
|
||||
"| item_id | item_name | qty |",
|
||||
"| :--- | :--- | ---: |",
|
||||
]
|
||||
for iid, (name, qty) in sorted(totals.items(), key=lambda x: x[0]):
|
||||
lines.append(
|
||||
"| {} | {} | {} |".format(_md_cell(iid), _md_cell(name), qty)
|
||||
)
|
||||
lines.append("")
|
||||
print_markdown_stderr("\n".join(lines))
|
||||
|
||||
|
||||
def append_consumption_window(
|
||||
*,
|
||||
surgery_id: str,
|
||||
@@ -215,9 +253,17 @@ def append_consumption_window(
|
||||
camera_id: str,
|
||||
wall_start_epoch: float,
|
||||
wall_end_epoch: float,
|
||||
running_totals: dict[str, tuple[str, int]] | None = None,
|
||||
) -> None:
|
||||
if not settings.consumption_tsv_log_enabled and not settings.consumption_log_markdown_terminal:
|
||||
return
|
||||
iid = resolve_consumption_item_id(best.t1_name, best.t1_pid, name_to_code)
|
||||
iname = (best.t1_name or "").strip()
|
||||
if running_totals is not None:
|
||||
if iid not in running_totals:
|
||||
running_totals[iid] = (iname, 0)
|
||||
prev_name, q = running_totals[iid]
|
||||
running_totals[iid] = (prev_name, q + 1)
|
||||
if settings.consumption_tsv_log_enabled:
|
||||
line = build_tsv_line(
|
||||
name_to_code=name_to_code,
|
||||
|
||||
@@ -2,6 +2,8 @@
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import base64
|
||||
|
||||
from app.database import AsyncSessionLocal
|
||||
from app.repositories.surgery_results import SurgeryResultRepository
|
||||
from app.schemas import (
|
||||
@@ -78,31 +80,22 @@ class SurgeryPipeline:
|
||||
return persisted
|
||||
return self._sessions.archived_consumption_fallback(surgery_id)
|
||||
|
||||
def voice_status(self, surgery_id: str) -> dict[str, object] | None:
|
||||
return self._sessions.voice_status(surgery_id)
|
||||
|
||||
async def list_voice_audits(
|
||||
self,
|
||||
surgery_id: str,
|
||||
*,
|
||||
limit: int = 50,
|
||||
offset: int = 0,
|
||||
):
|
||||
"""持久化表 `voice_confirmation_audits` 分页，用于追溯/对账/报表。"""
|
||||
return await self._voice.list_voice_audits_for_surgery(
|
||||
surgery_id, limit=limit, offset=offset
|
||||
)
|
||||
|
||||
def get_pending_confirmation_for_client(
|
||||
async def get_pending_confirmation_for_client(
|
||||
self, surgery_id: str
|
||||
) -> SurgeryPendingConfirmationResponse | None:
|
||||
pending = self._sessions.next_pending_confirmation(surgery_id)
|
||||
if pending is None:
|
||||
return None
|
||||
mp3 = await run_in_threadpool(
|
||||
self._voice.synthesize_prompt_to_mp3,
|
||||
pending.prompt_text,
|
||||
)
|
||||
b64 = base64.b64encode(mp3).decode("ascii")
|
||||
return SurgeryPendingConfirmationResponse(
|
||||
surgery_id=surgery_id,
|
||||
confirmation_id=pending.id,
|
||||
prompt_text=pending.prompt_text,
|
||||
prompt_audio_mp3_base64=b64,
|
||||
options=[
|
||||
PendingConfirmationOption(label=a, confidence=b)
|
||||
for a, b in pending.options
|
||||
@@ -129,34 +122,3 @@ class SurgeryPipeline:
|
||||
content_type=content_type,
|
||||
)
|
||||
|
||||
async def resolve_pending_confirmation_from_client_text(
|
||||
self,
|
||||
surgery_id: str,
|
||||
confirmation_id: str,
|
||||
recognized_text: str,
|
||||
) -> VoiceResolveResult:
|
||||
"""浏览器等客户端本机识别后的文本，解析规则与 WAV 路径一致（无需 MinIO/百度）。"""
|
||||
return await self._voice.resolve_from_recognized_text(
|
||||
surgery_id=surgery_id,
|
||||
confirmation_id=confirmation_id,
|
||||
recognized_text=recognized_text,
|
||||
)
|
||||
|
||||
async def get_pending_prompt_audio_mp3(
|
||||
self,
|
||||
surgery_id: str,
|
||||
confirmation_id: str,
|
||||
) -> bytes:
|
||||
"""待确认 `prompt_text` 的百度 TTS MP3，供模拟客户端用 Audio 直放。"""
|
||||
pending = self._sessions.get_pending_confirmation_by_id(
|
||||
surgery_id, confirmation_id
|
||||
)
|
||||
if pending is None or pending.status != "pending":
|
||||
raise SurgeryPipelineError(
|
||||
"CONFIRMATION_NOT_FOUND",
|
||||
"未找到该待确认项或已处理。",
|
||||
)
|
||||
return await run_in_threadpool(
|
||||
self._voice.synthesize_prompt_to_mp3,
|
||||
pending.prompt_text,
|
||||
)
|
||||
|
||||
@@ -12,7 +12,7 @@ from loguru import logger
|
||||
from app.config import Settings
|
||||
from app.database import AsyncSessionLocal
|
||||
from app.repositories.surgery_results import SurgeryResultRepository
|
||||
from app.schemas import SurgeryConsumptionDetail
|
||||
from app.schemas import SurgeryConsumptionDetail, SurgeryConsumptionStored
|
||||
from app.services.consumable_vision_algorithm import (
|
||||
ClsTop3,
|
||||
ConsumableVisionAlgorithmService,
|
||||
@@ -26,7 +26,12 @@ from app.services.video.backend_resolver import BackendResolver
|
||||
from app.services.video.hikvision_runtime import HikvisionInitRefCount, HikvisionRuntime
|
||||
from app.services.video.rtsp_capture import RtspCapture
|
||||
from app.services.video.types import VideoBackendKind
|
||||
from app.services.consumption_tsv_log import append_consumption_window, init_consumption_log_file
|
||||
from app.services.consumption_tsv_log import (
|
||||
append_consumption_log_summary,
|
||||
append_consumption_window,
|
||||
init_consumption_log_file,
|
||||
print_consumption_summary_markdown,
|
||||
)
|
||||
from app.services.voice_file_log import init_voice_log_file
|
||||
from app.services.voice_confirm import build_prompt_text
|
||||
from app.surgery_errors import SurgeryPipelineError
|
||||
@@ -64,7 +69,7 @@ class SurgerySessionState:
|
||||
#: 分类类名（归一化） -> 业务物品 id（Excel 产品编码或名称）。
|
||||
name_to_code: dict[str, str] = field(default_factory=dict)
|
||||
camera_infer: dict[str, CameraStreamInferState] = field(default_factory=dict)
|
||||
details: list[SurgeryConsumptionDetail] = field(default_factory=list)
|
||||
details: list[SurgeryConsumptionStored] = field(default_factory=list)
|
||||
lock: asyncio.Lock = field(default_factory=asyncio.Lock)
|
||||
ready: asyncio.Event = field(default_factory=asyncio.Event)
|
||||
last_detail_monotonic: dict[str, float] = field(default_factory=dict)
|
||||
@@ -76,6 +81,8 @@ class SurgerySessionState:
|
||||
last_asr_text: str | None = None
|
||||
#: 最近一次语音确认错误说明（ASR/解析失败等）。
|
||||
last_voice_error: str | None = None
|
||||
#: 视觉时间窗落盘用量累计，供停录时写汇总（item_id -> 首次名称, 次数）。
|
||||
consumption_log_totals: dict[str, tuple[str, int]] = field(default_factory=dict)
|
||||
|
||||
|
||||
@dataclass
|
||||
@@ -87,7 +94,7 @@ class RunningSurgery:
|
||||
|
||||
@dataclass
|
||||
class ArchivedSurgery:
|
||||
details: list[SurgeryConsumptionDetail]
|
||||
details: list[SurgeryConsumptionStored]
|
||||
|
||||
|
||||
def _rank_topk_for_candidates(
|
||||
@@ -293,7 +300,7 @@ class CameraSessionManager:
|
||||
async def _persist_archived_details(
|
||||
self,
|
||||
surgery_id: str,
|
||||
details: list[SurgeryConsumptionDetail],
|
||||
details: list[SurgeryConsumptionStored],
|
||||
) -> bool:
|
||||
if self._repo is None:
|
||||
return True
|
||||
@@ -331,6 +338,10 @@ class CameraSessionManager:
|
||||
if isinstance(res, BaseException):
|
||||
logger.warning("surgery task finished with error: {}", res)
|
||||
|
||||
totals = dict(run.state.consumption_log_totals)
|
||||
append_consumption_log_summary(surgery_id, totals)
|
||||
print_consumption_summary_markdown(totals)
|
||||
|
||||
details = list(run.state.details)
|
||||
|
||||
persisted = False
|
||||
@@ -364,26 +375,13 @@ class CameraSessionManager:
|
||||
rows = list(self._active[surgery_id].state.details)
|
||||
if not rows:
|
||||
return None
|
||||
return rows
|
||||
return [r.as_response() for r in rows]
|
||||
|
||||
def archived_consumption_fallback(self, surgery_id: str) -> list[SurgeryConsumptionDetail] | None:
|
||||
arch = self._archive.get(surgery_id)
|
||||
if arch is None:
|
||||
return None
|
||||
return list(arch.details)
|
||||
|
||||
def voice_status(self, surgery_id: str) -> dict[str, object] | None:
|
||||
if surgery_id not in self._active:
|
||||
return None
|
||||
st = self._active[surgery_id].state
|
||||
return {
|
||||
"surgery_id": surgery_id,
|
||||
"voice_enabled": bool(self._s.voice_confirmation_enabled),
|
||||
"pending_queue_approx": len(st.pending_fifo),
|
||||
"last_prompt_snippet": st.last_pending_prompt_snippet,
|
||||
"last_asr_text": st.last_asr_text,
|
||||
"last_error": st.last_voice_error,
|
||||
}
|
||||
return [r.as_response() for r in arch.details]
|
||||
|
||||
def record_voice_trace(
|
||||
self,
|
||||
@@ -525,10 +523,10 @@ class CameraSessionManager:
|
||||
return
|
||||
state.last_detail_monotonic[item_id] = now_m
|
||||
state.details.append(
|
||||
SurgeryConsumptionDetail(
|
||||
SurgeryConsumptionStored(
|
||||
item_id=item_id,
|
||||
item_name=item_name,
|
||||
quantity=1,
|
||||
qty=1,
|
||||
doctor_id=doctor_id,
|
||||
timestamp=datetime.now(timezone.utc),
|
||||
source=source,
|
||||
@@ -698,6 +696,7 @@ class CameraSessionManager:
|
||||
camera_id=camera_id,
|
||||
wall_start_epoch=wall_lo,
|
||||
wall_end_epoch=wall_hi,
|
||||
running_totals=state.consumption_log_totals,
|
||||
)
|
||||
pending_preds.append(
|
||||
cls_top3_to_prediction_result(best)
|
||||
|
||||
@@ -11,7 +11,6 @@ from loguru import logger
|
||||
from app.config import Settings
|
||||
from app.services.voice_file_log import emit_voice_event
|
||||
from app.database import AsyncSessionLocal
|
||||
from app.db.models import VoiceConfirmationAudit
|
||||
from app.repositories.voice_audits import VoiceAuditRepository
|
||||
from app.services.audio_wav import WavDecodeError, wav_bytes_to_pcm16k_mono_s16le
|
||||
from app.services.baidu_speech import BaiduSpeechNotConfiguredError, BaiduSpeechService
|
||||
@@ -660,22 +659,6 @@ class VoiceConfirmationService:
|
||||
message="已确认并记一条消耗。",
|
||||
)
|
||||
|
||||
async def list_voice_audits_for_surgery(
|
||||
self,
|
||||
surgery_id: str,
|
||||
*,
|
||||
limit: int = 50,
|
||||
offset: int = 0,
|
||||
) -> tuple[list[VoiceConfirmationAudit], int]:
|
||||
"""从 `voice_confirmation_audits` 表分页读取，供内部查询与报表。"""
|
||||
async with AsyncSessionLocal() as session:
|
||||
return await self._audits.list_by_surgery(
|
||||
session,
|
||||
surgery_id,
|
||||
limit=limit,
|
||||
offset=offset,
|
||||
)
|
||||
|
||||
async def _persist_audit(
|
||||
self,
|
||||
*,
|
||||
|
||||
Reference in New Issue
Block a user