feat: add voice transcription feedback workflow

.gitignore

@@ -3,6 +3,8 @@ graphify-out/
 
 # Dependencies and build output
 node_modules/
+.venv/
+.models/
 miniprogram_npm/
 dist/
 coverage/

DEBUG_GUIDE.md

@@ -0,0 +1,310 @@
+# 本地调试指南
+
+本文用于在本机启动 PostgreSQL 连接、中文 FunASR、Rust API 和微信小程序，并验证短录音、长录音及多课节汇总。
+
+## 1. 服务关系
+
+```text
+微信小程序
+  -> Rust API（127.0.0.1:8080）
+     -> PostgreSQL
+     -> FunASR（127.0.0.1:10095）
+```
+
+本地录音转写的核心是 FunASR，当前组合为：
+
+- Paraformer：中文语音识别。
+- FSMN-VAD：检测并切分有效语音。
+- CT-Transformer：恢复中文标点。
+- PyTorch：CPU 推理运行时。
+
+Rust API 负责音频保存、持久化任务队列、重试、课节状态和反馈汇总。小程序每 8 分钟自动切片，并每 3 秒无感刷新转写状态。
+
+## 2. 当前主机资源实测
+
+测试主机为 Apple Silicon、10 核 CPU、16 GB 统一内存，配置为 CPU 单并发：
+
+| 项目 | 实测值 |
+| --- | --- |
+| FunASR 模型缓存 | 约 2.1 GB |
+| Python 虚拟环境 | 约 871 MB |
+| 模型加载后的物理内存 | 约 2.8 GB |
+| 进程历史内存峰值 | 约 4.9 GB |
+| 8 分钟、16 kHz、单声道、48 kbps MP3 | 约 44.8 秒完成 |
+| 推理期间 CPU | 约 90% 至 132%，即约 1 至 1.3 个 CPU 核 |
+
+8 分钟数据使用重复的清晰中文样本测得，真实课堂中的噪声、停顿和说话人数会影响耗时和识别质量。
+
+容量建议：
+
+- 当前单教师或低并发调试：不需要 GPU，当前 16 GB 主机足够。
+- CPU 部署最低建议：4 vCPU、8 GB 内存；8 GB 可能发生交换，16 GB 更稳妥。
+- 多教师并发：先保持单并发观察队列；若片段持续积压，再考虑增加 CPU 实例或 NVIDIA GPU。
+- GPU 是可选优化，不是运行前提。CUDA 部署需要单独的 CUDA/PyTorch 镜像，当前 CPU 镜像不能只修改 `ASR_DEVICE=cuda`。
+- GPU 部署建议从 8 GB 显存起步并用真实课堂录音压测。当前 Apple Silicon 配置没有使用 Apple GPU。
+- `ASR_WORKER_CONCURRENCY` 和 `ASR_MAX_CONCURRENCY` 应保持一致；每提高一个并发都要重新观察内存和吞吐。
+
+## 3. 首次准备
+
+在项目根目录执行：
+
+```bash
+cd /Users/zhangheng/project/ai/miniprogram/teaching-feedback-assistant
+```
+
+确认 `server/.env` 存在，并填写可访问的 `DATABASE_URL`。不存在时再复制示例，避免覆盖现有配置：
+
+```bash
+test -f server/.env || cp server/.env.example server/.env
+```
+
+本地 FunASR 默认配置如下，无需腾讯云密钥：
+
+```env
+ASR_PROVIDER=local
+LOCAL_ASR_URL=http://127.0.0.1:10095
+ASR_REQUEST_TIMEOUT_SECONDS=1800
+ASR_WORKER_CONCURRENCY=1
+ASR_JOB_LEASE_SECONDS=3600
+```
+
+首次准备 Python 环境：
+
+```bash
+cd asr-service
+python3 -m venv .venv
+.venv/bin/python -m pip install --upgrade pip
+.venv/bin/python -m pip install -r requirements.txt
+cd ..
+```
+
+模型首次启动会下载约 2.1 GB 到 `asr-service/.models`。`.venv` 和 `.models` 均已被 Git 和微信小程序打包忽略。
+
+## 4. 启动全部服务
+
+### 4.1 执行数据库迁移
+
+每次拉取到新迁移后执行一次：
+
+```bash
+cd server
+cargo run --bin migrate
+cd ..
+```
+
+看到 `PostgreSQL migrations completed` 表示成功。
+
+### 4.2 启动 FunASR
+
+终端一：
+
+```bash
+cd asr-service
+MODELSCOPE_CACHE=.models .venv/bin/uvicorn app:app \
+  --host 127.0.0.1 \
+  --port 10095 \
+  --no-access-log
+```
+
+另一个终端检查：
+
+```bash
+curl http://127.0.0.1:10095/health
+```
+
+预期结果：
+
+```json
+{"status":"ok","model":"paraformer-zh","device":"cpu"}
+```
+
+首次下载或缓存加载期间端口可能暂不可用，应等待模型完全加载。
+
+Docker 方式可替代本机 Python：
+
+```bash
+docker compose -f compose.asr.yml up --build -d
+docker compose -f compose.asr.yml logs -f funasr
+```
+
+### 4.3 启动 Rust API
+
+终端二：
+
+```bash
+cd server
+cargo run
+```
+
+检查完整链路：
+
+```bash
+curl http://127.0.0.1:8080/health | jq .
+```
+
+正常状态应包含：
+
+```json
+{
+  "status": "ok",
+  "database_configured": true,
+  "speech_configured": true,
+  "speech_available": true,
+  "speech_provider": "local-funasr"
+}
+```
+
+接口文档：<http://127.0.0.1:8080/scalar>
+
+### 4.4 启动微信开发者工具
+
+终端三，在项目根目录执行：
+
+```bash
+cli open \
+  --project /Users/zhangheng/project/ai/miniprogram/teaching-feedback-assistant \
+  --port 63097 \
+  --lang zh
+```
+
+自动预览编译检查：
+
+```bash
+cli auto-preview \
+  --project /Users/zhangheng/project/ai/miniprogram/teaching-feedback-assistant \
+  --port 63097 \
+  --lang zh \
+  --info-output /tmp/teaching-feedback-preview.json
+```
+
+开发者工具模拟器默认访问 `http://127.0.0.1:8080`。进入小程序“我的”页面，应看到“服务、数据库和语音转录正常”。
+
+## 5. 录音测试流程
+
+### 短录音
+
+1. 打开“反馈生成”。
+2. 点击“语音录入”，录制 10 至 30 秒中文。
+3. 点击“结束录音”。
+4. 页面短暂显示“正在处理”。
+5. 转写完成后文字应自动加入反馈内容。
+
+短录音条件为不超过 60 秒，且转写文本未超过后端短文本限制。
+
+### 长录音
+
+1. 录制超过 60 秒的中文。
+2. 结束后等待状态变成“已就绪”。
+3. 主按钮变为“生成反馈”。
+4. 点击一次后，转写内容会汇总进反馈草稿。
+
+### 多课节汇总
+
+1. 录制并结束第一节课。
+2. 再次点击“语音录入”录制下一节。
+3. 重复完成多节录音。
+4. 所有课节就绪后点击一次“生成反馈”。
+
+单节录音超过 8 分钟时，小程序会自动结束当前片段并立即开始下一片段，用户无需操作。
+
+## 6. 故障恢复测试
+
+### FunASR 未启动
+
+停止 FunASR 后录音，Rust API 仍会保存音频，页面保持“正在处理”。重新启动 FunASR 后，后台队列会自动继续。
+
+### Rust 在推理中退出
+
+任务使用租约避免多个工作线程重复处理。默认租约为 3600 秒。调试崩溃恢复时，可在 `server/.env` 临时设置：
+
+```env
+ASR_JOB_LEASE_SECONDS=60
+```
+
+重新启动 Rust API 后，超出租约的任务会再次被领取。生产环境应使用足够覆盖最长片段推理时间的租约。
+
+### 转写失败
+
+点击页面中的语音状态，可查看课节状态并重试失败片段。音频保存在 `server/data/audio`，转写失败不会立即删除原文件。
+
+## 7. 常见问题
+
+### `speech_configured=true` 但 `speech_available=false`
+
+- FunASR 尚未启动或模型仍在加载。
+- `10095` 端口被占用。
+- `LOCAL_ASR_URL` 配置错误。
+
+检查：
+
+```bash
+lsof -nP -iTCP:10095 -sTCP:LISTEN
+curl http://127.0.0.1:10095/health
+```
+
+### `database_configured=false` 或数据接口返回 503
+
+检查 `server/.env` 中的 `DATABASE_URL`，然后重新执行迁移并启动 API。
+
+### 微信模拟器无法连接 API
+
+```bash
+lsof -nP -iTCP:8080 -sTCP:LISTEN
+curl http://127.0.0.1:8080/health
+```
+
+再到小程序“我的”页面确认 API 地址为 `http://127.0.0.1:8080`。
+
+### 真机无法访问 `127.0.0.1`
+
+真机中的 `127.0.0.1` 指手机自身。局域网调试时：
+
+1. 将 `server/.env` 的 `HOST` 改为 `0.0.0.0`。
+2. 确认手机和电脑连接同一网络。
+3. 在“我的”页面将 API 地址改为 `http://<电脑局域网IP>:8080`。
+4. 检查系统防火墙和微信开发者工具的域名校验设置。
+
+正式环境必须使用微信允许的 HTTPS 合法域名。
+
+## 8. 代码检查
+
+小程序：
+
+```bash
+npm run typecheck
+```
+
+Rust：
+
+```bash
+cd server
+cargo fmt --check
+cargo test
+cargo clippy --all-targets -- -D warnings
+```
+
+FunASR 服务：
+
+```bash
+python3 -m py_compile asr-service/app.py
+asr-service/.venv/bin/python -m pip check
+```
+
+## 9. 停止服务
+
+前台运行时，在 FunASR 和 Rust API 各自终端按 `Ctrl+C`。
+
+Docker 方式：
+
+```bash
+docker compose -f compose.asr.yml down
+```
+
+确认端口已经释放：
+
+```bash
+lsof -nP -iTCP:8080 -sTCP:LISTEN
+lsof -nP -iTCP:10095 -sTCP:LISTEN
+```
+
+正常停止服务不会删除 `asr-service/.models` 模型缓存。除非希望重新下载模型，否则不要删除该目录。

README.md

@@ -11,6 +11,8 @@ cli open --project /Users/zhangheng/project/ai/miniprogram/teaching-feedback-ass
 
 3. 在微信开发者工具内编译、预览和调试。
 
+反馈生成页使用小程序原生录音能力。短录音会直接追加到反馈内容，长录音每 8 分钟自动切片，多节录音可以汇总为一次反馈。真机首次使用时需要允许麦克风权限；中文转录默认使用项目自带的本地 FunASR 服务，不要求腾讯云密钥。
+
 小程序默认连接 `http://127.0.0.1:8080`。在开发者工具模拟器中，可进入“我的”页面保存其他 API 地址并检测服务状态。
 
 ## 验证
@@ -33,6 +35,15 @@ cargo run --bin migrate
 cargo run
 ```
 
+另开一个终端启动中文转录服务。首次启动会下载 Paraformer、VAD 和标点模型，下载完成后健康检查才会显示语音服务可用：
+
+```bash
+docker compose -f compose.asr.yml up --build -d
+curl http://127.0.0.1:10095/health
+```
+
+没有 Docker 时可按 [asr-service/README.md](./asr-service/README.md) 使用 Python 3.10 启动。Rust API 默认连接 `http://127.0.0.1:10095`；模型未就绪时已上传录音会保留在队列中，不会阻塞小程序上传。
+
 未设置 `DATABASE_URL` 时，仅健康检查可用，数据接口会返回 `503`。数据库配置和迁移说明参见 [server/API.md](./server/API.md)。
 
 服务启动后可打开 `http://127.0.0.1:8080/scalar` 查看并直接调用 Scalar 交互文档，原始规范位于 `http://127.0.0.1:8080/openapi.json`。完整测试顺序参见 [server/API_GUIDE.md](./server/API_GUIDE.md)。
@@ -40,6 +51,7 @@ cargo run
 ## 小程序与后端
 
 - 学生档案、档案预设和反馈记录均通过 `utils/api.ts` 访问 Rust API。
+- 反馈页只暴露一个语音入口；反馈批次、课节、录音片段和转录状态由后端自动维护。
 - 开发者工具模拟器使用默认地址 `http://127.0.0.1:8080`。
 - 真机和正式版本不能把 `127.0.0.1` 当作电脑地址；应部署可公网访问的 HTTPS API，并在微信公众平台配置 request 合法域名，然后到小程序“我的”页面修改服务地址。
 - 微信登录尚未接入。业务请求暂时使用固定开发用户 UUID；生产发布前必须改为由后端根据微信登录凭据解析用户身份。

app.json

@@ -13,6 +13,11 @@
     "backgroundColor": "#F4F8FE",
     "backgroundTextStyle": "light"
   },
+  "permission": {
+    "scope.record": {
+      "desc": "用于将课堂反馈录音转成文字"
+    }
+  },
   "tabBar": {
     "custom": true,
     "color": "#7B8796",

asr-service/.dockerignore

@@ -0,0 +1,4 @@
+__pycache__/
+*.pyc
+.venv/
+.pytest_cache/

asr-service/Dockerfile

@@ -0,0 +1,20 @@
+FROM python:3.10-slim
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    PIP_NO_CACHE_DIR=1
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends ffmpeg libsndfile1 \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+COPY requirements.txt ./
+RUN pip install --upgrade pip \
+    && pip install torch==2.8.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/cpu \
+    && pip install -r requirements.txt
+
+COPY app.py ./
+
+EXPOSE 10095
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "10095", "--workers", "1", "--no-access-log"]

asr-service/README.md

@@ -0,0 +1,38 @@
+# 中文录音转写服务
+
+该服务使用 FunASR 的中文 Paraformer、FSMN-VAD 和标点模型，为 Rust API 提供本地录音转写。模型首次启动时会下载约 2.1 GB 到持久化缓存，完成后 `/health` 才会可用。
+
+## Docker 启动
+
+在项目根目录执行：
+
+```bash
+docker compose -f compose.asr.yml up --build -d
+curl http://127.0.0.1:10095/health
+```
+
+默认使用 CPU 且只允许一个并发推理任务。可通过 `ASR_DEVICE`、`ASR_MAX_CONCURRENCY`、`ASR_MODEL`、`ASR_VAD_MODEL` 和 `ASR_PUNC_MODEL` 调整。GPU 部署需要使用带 CUDA 的 PyTorch 基础环境，不能只把 CPU 镜像中的 `ASR_DEVICE` 改为 `cuda`。
+
+## 本机 Python 启动
+
+支持 Python 3.9 及以上版本，推荐 Python 3.10：
+
+```bash
+cd asr-service
+python3.10 -m venv .venv
+source .venv/bin/activate
+pip install -r requirements.txt
+MODELSCOPE_CACHE=.models uvicorn app:app --host 127.0.0.1 --port 10095 --no-access-log
+```
+
+## 接口
+
+Rust API 调用 `POST /v1/transcriptions`，以 multipart 上传 `file`、`language` 和 `duration_ms`，响应格式为：
+
+```json
+{
+  "text": "识别后的中文课堂内容。",
+  "duration_ms": 480000,
+  "request_id": "b2f6fc1c-62c1-4a99-8ab4-6f933f1bd252"
+}
+```

asr-service/app.py

@@ -0,0 +1,173 @@
+from __future__ import annotations
+
+import asyncio
+import logging
+import os
+import tempfile
+import uuid
+from contextlib import asynccontextmanager
+from pathlib import Path
+from typing import Any
+
+from fastapi import FastAPI, File, Form, HTTPException, UploadFile
+from funasr import AutoModel
+from pydantic import BaseModel
+
+LOGGER = logging.getLogger("funasr-service")
+MAX_AUDIO_BYTES = 15 * 1024 * 1024
+CHUNK_BYTES = 1024 * 1024
+SUPPORTED_FORMATS = {"mp3", "aac", "wav", "m4a", "amr"}
+
+
+class Settings:
+    def __init__(self) -> None:
+        self.model = os.getenv("ASR_MODEL", "paraformer-zh")
+        self.vad_model = os.getenv("ASR_VAD_MODEL", "fsmn-vad")
+        self.punc_model = os.getenv("ASR_PUNC_MODEL", "ct-punc")
+        self.device = os.getenv("ASR_DEVICE", "cpu")
+        self.max_concurrency = positive_int("ASR_MAX_CONCURRENCY", 1)
+        self.batch_size_seconds = positive_int("ASR_BATCH_SIZE_SECONDS", 300)
+
+
+class HealthResponse(BaseModel):
+    status: str
+    model: str
+    device: str
+
+
+class TranscriptionResponse(BaseModel):
+    text: str
+    duration_ms: int
+    request_id: str
+
+
+def positive_int(name: str, default: int) -> int:
+    raw = os.getenv(name, str(default))
+    try:
+        value = int(raw)
+    except ValueError as error:
+        raise RuntimeError(f"{name} must be an integer") from error
+    if value <= 0:
+        raise RuntimeError(f"{name} must be greater than 0")
+    return value
+
+
+def load_model(settings: Settings) -> Any:
+    LOGGER.info(
+        "loading FunASR model model=%s vad=%s punc=%s device=%s",
+        settings.model,
+        settings.vad_model,
+        settings.punc_model,
+        settings.device,
+    )
+    model = AutoModel(
+        model=settings.model,
+        vad_model=settings.vad_model,
+        punc_model=settings.punc_model,
+        device=settings.device,
+        disable_update=True,
+    )
+    LOGGER.info("FunASR model is ready")
+    return model
+
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    settings = Settings()
+    app.state.settings = settings
+    app.state.model = await asyncio.to_thread(load_model, settings)
+    app.state.inference_slots = asyncio.Semaphore(settings.max_concurrency)
+    yield
+
+
+app = FastAPI(
+    title="Teaching Feedback FunASR Service",
+    version="1.0.0",
+    lifespan=lifespan,
+)
+
+
+@app.get("/health", response_model=HealthResponse)
+async def health() -> HealthResponse:
+    settings: Settings = app.state.settings
+    return HealthResponse(status="ok", model=settings.model, device=settings.device)
+
+
+@app.post("/v1/transcriptions", response_model=TranscriptionResponse)
+async def create_transcription(
+    file: UploadFile = File(...),
+    language: str = Form("zh"),
+    duration_ms: int = Form(0),
+    hotword: str = Form(""),
+) -> TranscriptionResponse:
+    if language not in {"zh", "zh-cn", "cmn"}:
+        raise HTTPException(status_code=422, detail="this service is configured for Chinese audio")
+    audio_format = file_extension(file.filename)
+    temporary_path = await save_upload(file, audio_format)
+    request_id = str(uuid.uuid4())
+    try:
+        async with app.state.inference_slots:
+            text = await asyncio.to_thread(
+                run_inference,
+                app.state.model,
+                temporary_path,
+                app.state.settings.batch_size_seconds,
+                hotword.strip(),
+            )
+    except Exception as error:
+        LOGGER.exception("FunASR inference failed request_id=%s", request_id)
+        raise HTTPException(status_code=500, detail=f"speech recognition failed: {error}") from error
+    finally:
+        temporary_path.unlink(missing_ok=True)
+
+    if not text:
+        raise HTTPException(status_code=422, detail="no clear speech was recognized")
+    return TranscriptionResponse(
+        text=text,
+        duration_ms=max(0, duration_ms),
+        request_id=request_id,
+    )
+
+
+def file_extension(filename: str | None) -> str:
+    suffix = Path(filename or "segment.mp3").suffix.lower().lstrip(".")
+    if suffix not in SUPPORTED_FORMATS:
+        raise HTTPException(status_code=415, detail="unsupported audio format")
+    return suffix
+
+
+async def save_upload(file: UploadFile, audio_format: str) -> Path:
+    handle = tempfile.NamedTemporaryFile(suffix=f".{audio_format}", delete=False)
+    path = Path(handle.name)
+    total_bytes = 0
+    try:
+        with handle:
+            while chunk := await file.read(CHUNK_BYTES):
+                total_bytes += len(chunk)
+                if total_bytes > MAX_AUDIO_BYTES:
+                    raise HTTPException(status_code=413, detail="audio file exceeds 15 MB")
+                handle.write(chunk)
+    except Exception:
+        path.unlink(missing_ok=True)
+        raise
+    finally:
+        await file.close()
+    if total_bytes == 0:
+        path.unlink(missing_ok=True)
+        raise HTTPException(status_code=422, detail="audio file is empty")
+    return path
+
+
+def run_inference(model: Any, path: Path, batch_size_seconds: int, hotword: str) -> str:
+    options: dict[str, Any] = {
+        "input": str(path),
+        "batch_size_s": batch_size_seconds,
+        "use_itn": True,
+    }
+    if hotword:
+        options["hotword"] = hotword
+    result = model.generate(**options)
+    if not isinstance(result, list):
+        return ""
+    texts = [str(item.get("text", "")).strip() for item in result if isinstance(item, dict)]
+    return "\n".join(text for text in texts if text).strip()

asr-service/requirements.txt

@@ -0,0 +1,7 @@
+fastapi==0.128.8
+funasr==1.3.14
+modelscope==1.37.1
+python-multipart==0.0.20
+torch==2.8.0
+torchaudio==2.8.0
+uvicorn[standard]==0.39.0

compose.asr.yml

@@ -0,0 +1,31 @@
+services:
+  funasr:
+    build:
+      context: ./asr-service
+    ports:
+      - "127.0.0.1:10095:10095"
+    environment:
+      ASR_MODEL: paraformer-zh
+      ASR_VAD_MODEL: fsmn-vad
+      ASR_PUNC_MODEL: ct-punc
+      ASR_DEVICE: cpu
+      ASR_MAX_CONCURRENCY: "1"
+      ASR_BATCH_SIZE_SECONDS: "300"
+      MODELSCOPE_CACHE: /models
+    volumes:
+      - funasr-model-cache:/models
+    restart: unless-stopped
+    healthcheck:
+      test:
+        - CMD
+        - python
+        - -c
+        - "import urllib.request; urllib.request.urlopen('http://127.0.0.1:10095/health', timeout=3)"
+      interval: 15s
+      timeout: 5s
+      retries: 20
+      start_period: 5m
+    shm_size: 1gb
+
+volumes:
+  funasr-model-cache:

pages/feedback/index.ts

@@ -1,13 +1,54 @@
 import {
   createFeedbackRecord,
+  createVoiceLesson,
+  ensureFeedbackSession,
+  finishVoiceLesson,
+  generateFeedbackFromVoice,
+  getActiveFeedbackSession,
   getApiErrorMessage,
-  listProfiles
+  listProfiles,
+  markVoiceLessonApplied,
+  retryVoiceSegment,
+  updateFeedbackSession,
+  uploadVoiceSegment
 } from '../../utils/api'
-import type { FeedbackDraft, StudentProfile } from '../../utils/types'
+import type {
+  FeedbackDraft,
+  FeedbackSession,
+  StudentProfile,
+  VoiceLessonFinished
+} from '../../utils/types'
 
 type PickerEvent = { detail: { value: string | number } }
 type InputEvent = { detail: { value: string } }
 
+interface PendingVoiceUpload {
+  lessonId: string
+  sequence: number
+  durationMs: number
+  filePath: string
+}
+
+const MAX_CONTENT_LENGTH = 2000
+const RECORDING_SEGMENT_DURATION_MS = 8 * 60 * 1000
+const SHORT_RECORDING_DURATION_MS = 60 * 1000
+const VOICE_POLL_INTERVAL_MS = 3000
+const PENDING_UPLOADS_KEY = 'teaching-feedback-pending-voice-uploads-v1'
+
+const recorderManager = wx.getRecorderManager()
+let recordingTimer: number | null = null
+let draftSaveTimer: number | null = null
+let voicePollTimer: number | null = null
+let voicePollBusy = false
+let pageVisible = false
+let lessonStartedAt = 0
+let segmentStartedAt = 0
+let currentLessonId: string | null = null
+let currentSegmentSequence = 0
+let finishRequested = false
+let pendingUploads: Promise<void>[] = []
+let failedLocalUploads: PendingVoiceUpload[] = []
+
 function today(): string {
   const now = new Date()
   const year = now.getFullYear()
@@ -20,39 +61,147 @@ function buildDraft(): FeedbackDraft {
   return {
     profileId: null,
     feedbackDate: today(),
-    content: ''
+    content: '',
+    feedbackSessionId: null
   }
 }
 
+function appendTranscript(content: string, transcript: string): { content: string; truncated: boolean } {
+  const normalizedTranscript = transcript.trim()
+  const separator = content && !/\s$/.test(content) ? '\n' : ''
+  const combined = `${content}${separator}${normalizedTranscript}`
+  return {
+    content: combined.slice(0, MAX_CONTENT_LENGTH),
+    truncated: combined.length > MAX_CONTENT_LENGTH
+  }
+}
+
+function formatRecordingTime(milliseconds: number): string {
+  const totalSeconds = Math.max(0, Math.floor(milliseconds / 1000))
+  const hours = Math.floor(totalSeconds / 3600)
+  const minutes = Math.floor((totalSeconds % 3600) / 60)
+  const seconds = totalSeconds % 60
+  const minuteText = String(minutes).padStart(2, '0')
+  const secondText = String(seconds).padStart(2, '0')
+  return hours > 0 ? `${String(hours).padStart(2, '0')}:${minuteText}:${secondText}` : `${minuteText}:${secondText}`
+}
+
+function formatVoiceDuration(milliseconds: number): string {
+  const totalMinutes = Math.floor(milliseconds / 60000)
+  if (totalMinutes < 1) return `${Math.max(1, Math.round(milliseconds / 1000))}秒`
+  return `${totalMinutes}分钟`
+}
+
+function wait(milliseconds: number): Promise<void> {
+  return new Promise((resolve) => setTimeout(resolve, milliseconds))
+}
+
 Page({
   data: {
     profiles: [] as StudentProfile[],
     profileOptions: ['不关联学生'],
     selectedProfileIndex: 0,
     draft: buildDraft(),
+    session: null as FeedbackSession | null,
     contentLength: 0,
     loading: false,
     saving: false,
+    generating: false,
+    voiceBusy: false,
+    recording: false,
+    recordingPaused: false,
+    recordingTime: '00:00',
+    voiceStatus: '语音录入',
+    voiceActionable: false,
+    primaryActionText: '保存反馈',
     errorMessage: ''
   },
 
+  onLoad() {
+    failedLocalUploads = (wx.getStorageSync(PENDING_UPLOADS_KEY) as PendingVoiceUpload[]) || []
+    this.initRecorder()
+    if (failedLocalUploads.length > 0) {
+      currentLessonId = failedLocalUploads[0].lessonId
+      this.setData({ voiceStatus: '上传失败，点击重试', voiceActionable: true })
+    }
+  },
+
   onShow() {
+    pageVisible = true
     this.getTabBar()?.setData({ selected: 0 })
     void this.loadProfiles()
   },
 
+  onHide() {
+    pageVisible = false
+    this.clearVoicePollTimer()
+    if (this.data.recording) this.stopRecording()
+  },
+
+  onUnload() {
+    pageVisible = false
+    this.clearRecordingTimer()
+    this.clearDraftSaveTimer()
+    this.clearVoicePollTimer()
+    if (this.data.recording) recorderManager.stop()
+  },
+
+  initRecorder() {
+    recorderManager.onStart(() => {
+      segmentStartedAt = Date.now()
+      this.setData({
+        voiceBusy: false,
+        recording: true,
+        recordingPaused: false,
+        voiceStatus: '正在录音'
+      })
+      this.startRecordingTimer()
+    })
+
+    recorderManager.onStop((result) => {
+      void this.handleRecorderStop(result)
+    })
+
+    recorderManager.onPause(() => {
+      this.setData({ recordingPaused: true, voiceStatus: '录音已暂停' })
+    })
+
+    recorderManager.onResume(() => {
+      this.setData({ recordingPaused: false, voiceStatus: '正在录音' })
+    })
+
+    recorderManager.onInterruptionEnd(() => {
+      if (this.data.recordingPaused) recorderManager.resume()
+    })
+
+    recorderManager.onError((error) => {
+      console.warn('RecorderManager error', error)
+      this.clearRecordingTimer()
+      finishRequested = true
+      this.setData({
+        recording: false,
+        recordingPaused: false,
+        voiceBusy: true,
+        voiceStatus: '正在保留已录内容'
+      })
+      void this.finalizeCurrentLesson()
+    })
+  },
+
   async loadProfiles() {
     this.setData({ loading: true, errorMessage: '' })
     try {
       const profiles = await listProfiles()
       const matchedProfileIndex = profiles.findIndex((profile) => profile.id === this.data.draft.profileId)
       const selectedProfileIndex = matchedProfileIndex + 1
+      const profileId = matchedProfileIndex >= 0 ? profiles[matchedProfileIndex].id : null
       this.setData({
         profiles,
         profileOptions: ['不关联学生', ...profiles.map((profile) => `${profile.name} · ${profile.subject}`)],
         selectedProfileIndex,
-        'draft.profileId': matchedProfileIndex >= 0 ? profiles[matchedProfileIndex].id : null
+        'draft.profileId': profileId
       })
+      await this.loadActiveSession(profileId)
     } catch (error) {
       this.setData({ errorMessage: getApiErrorMessage(error) })
     } finally {
@@ -60,29 +209,483 @@ Page({
     }
   },
 
+  async loadActiveSession(profileId?: string | null) {
+    try {
+      const targetProfileId = profileId === undefined ? this.data.draft.profileId : profileId
+      const session = await getActiveFeedbackSession(targetProfileId)
+      if (targetProfileId !== this.data.draft.profileId) return
+      const nextData: Record<string, unknown> = {
+        session,
+        'draft.feedbackSessionId': session?.id || null
+      }
+      if (session) {
+        nextData['draft.feedbackDate'] = session.feedbackDate
+        nextData['draft.content'] = session.content
+        nextData.contentLength = session.content.length
+      }
+      this.setData(nextData)
+      this.syncVoicePresentation(session)
+      this.scheduleVoicePolling(session)
+    } catch (error) {
+      this.setData({ errorMessage: getApiErrorMessage(error) })
+    }
+  },
+
+  syncVoicePresentation(session: FeedbackSession | null) {
+    if (this.data.recording || this.data.voiceBusy) return
+    if (failedLocalUploads.length > 0) {
+      this.setData({ voiceStatus: '上传失败，点击重试', voiceActionable: true })
+      return
+    }
+    if (!session || session.lessonCount === 0) {
+      this.setData({ voiceStatus: '语音录入', voiceActionable: false, primaryActionText: '保存反馈' })
+      return
+    }
+
+    const prefix = `${session.lessonCount}节 · ${formatVoiceDuration(session.totalDurationMs)}`
+    let suffix = '已就绪'
+    if (session.failedSegmentCount > 0) suffix = '转录失败，点击重试'
+    else if (session.processingSegmentCount > 0) suffix = '正在处理'
+    let primaryActionText = session.needsGeneration ? '生成反馈' : '保存反馈'
+    if (session.processingSegmentCount > 0) primaryActionText = '转录中'
+    this.setData({
+      voiceStatus: `${prefix} · ${suffix}`,
+      voiceActionable: true,
+      primaryActionText
+    })
+  },
+
+  scheduleVoicePolling(session: FeedbackSession | null) {
+    this.clearVoicePollTimer()
+    if (!pageVisible || !session || session.processingSegmentCount === 0) return
+    voicePollTimer = setTimeout(() => {
+      voicePollTimer = null
+      void this.pollVoiceSession()
+    }, VOICE_POLL_INTERVAL_MS) as unknown as number
+  },
+
+  clearVoicePollTimer() {
+    if (voicePollTimer === null) return
+    clearTimeout(voicePollTimer)
+    voicePollTimer = null
+  },
+
+  async pollVoiceSession() {
+    if (!pageVisible) return
+    if (voicePollBusy || this.data.recording || this.data.voiceBusy) {
+      this.scheduleVoicePolling(this.data.session)
+      return
+    }
+    voicePollBusy = true
+    const profileId = this.data.draft.profileId
+    try {
+      const session = await getActiveFeedbackSession(profileId)
+      if (!pageVisible || profileId !== this.data.draft.profileId) return
+      this.setData({ session, 'draft.feedbackSessionId': session?.id || null })
+      this.syncVoicePresentation(session)
+      if (!session) return
+
+      const readyShortLessons = session.lessons.filter(
+        (lesson) =>
+          lesson.status === 'ready' &&
+          lesson.durationMs <= SHORT_RECORDING_DURATION_MS &&
+          !lesson.appliedDirectly &&
+          !lesson.includedInGeneration
+      )
+      for (const lesson of readyShortLessons) {
+        const finished = await finishVoiceLesson(lesson.id)
+        await this.applyFinishedLesson(finished)
+      }
+      if (readyShortLessons.length > 0) await this.loadActiveSession(profileId)
+    } catch (error) {
+      console.warn('Failed to refresh voice transcription status', error)
+    } finally {
+      voicePollBusy = false
+      this.scheduleVoicePolling(this.data.session)
+    }
+  },
+
   retryLoad() {
     void this.loadProfiles()
   },
 
   onProfileChange(event: PickerEvent) {
+    this.clearDraftSaveTimer()
     const selectedProfileIndex = Number(event.detail.value)
     const profile = this.data.profiles[selectedProfileIndex - 1]
+    const profileId = profile?.id || null
     this.setData({
       selectedProfileIndex,
-      'draft.profileId': profile?.id || null
+      'draft.profileId': profileId,
+      session: null,
+      'draft.feedbackSessionId': null,
+      'draft.content': '',
+      contentLength: 0,
+      voiceActionable: false,
+      primaryActionText: '保存反馈'
     })
+    void this.loadActiveSession(profileId)
   },
 
   onDateChange(event: InputEvent) {
     this.setData({ 'draft.feedbackDate': event.detail.value })
+    this.scheduleDraftSave()
   },
 
   onContentInput(event: InputEvent) {
     const content = event.detail.value
     this.setData({ 'draft.content': content, contentLength: content.length })
+    this.scheduleDraftSave()
+  },
+
+  scheduleDraftSave() {
+    this.clearDraftSaveTimer()
+    if (!this.data.session) return
+    draftSaveTimer = setTimeout(() => {
+      void this.persistSessionDraft()
+    }, 800) as unknown as number
+  },
+
+  clearDraftSaveTimer() {
+    if (draftSaveTimer === null) return
+    clearTimeout(draftSaveTimer)
+    draftSaveTimer = null
+  },
+
+  async persistSessionDraft() {
+    const session = this.data.session
+    if (!session) return
+    const sessionId = session.id
+    try {
+      const updated = await updateFeedbackSession(
+        sessionId,
+        this.data.draft.feedbackDate,
+        this.data.draft.content
+      )
+      if (this.data.session?.id !== sessionId) return
+      this.setData({ session: updated })
+      this.syncVoicePresentation(updated)
+    } catch (error) {
+      console.warn('Failed to auto-save feedback session', error)
+    }
+  },
+
+  ensureRecordPermission(): Promise<boolean> {
+    return new Promise((resolve) => {
+      wx.authorize({
+        scope: 'scope.record',
+        success: () => resolve(true),
+        fail: () => {
+          wx.showModal({
+            title: '需要麦克风权限',
+            content: '请在设置中允许使用麦克风，才能录音转录。',
+            confirmText: '去设置',
+            success: (modalResult) => {
+              if (!modalResult.confirm) {
+                resolve(false)
+                return
+              }
+              wx.openSetting({
+                success: (settingResult) => resolve(settingResult.authSetting['scope.record'] === true),
+                fail: () => resolve(false)
+              })
+            },
+            fail: () => resolve(false)
+          })
+        }
+      })
+    })
+  },
+
+  async toggleRecording() {
+    if (failedLocalUploads.length > 0) {
+      await this.retryPendingUploads()
+      return
+    }
+    if (this.data.recording) {
+      this.stopRecording()
+      return
+    }
+
+    this.setData({ voiceBusy: true, voiceStatus: '正在开启麦克风' })
+    const authorized = await this.ensureRecordPermission()
+    if (!authorized) {
+      this.setData({ voiceBusy: false })
+      this.syncVoicePresentation(this.data.session)
+      return
+    }
+
+    try {
+      const session = await ensureFeedbackSession(
+        this.data.draft.profileId,
+        this.data.draft.feedbackDate,
+        this.data.draft.content
+      )
+      const lesson = await createVoiceLesson(session.id)
+      currentLessonId = lesson.id
+      currentSegmentSequence = 0
+      finishRequested = false
+      pendingUploads = []
+      lessonStartedAt = Date.now()
+      this.setData({
+        session,
+        'draft.feedbackSessionId': session.id,
+        recordingTime: '00:00'
+      })
+      this.startNextRecordingSegment()
+    } catch (error) {
+      this.setData({ voiceBusy: false })
+      this.syncVoicePresentation(this.data.session)
+      wx.showToast({ title: getApiErrorMessage(error), icon: 'none' })
+    }
+  },
+
+  startNextRecordingSegment() {
+    currentSegmentSequence += 1
+    recorderManager.start({
+      duration: RECORDING_SEGMENT_DURATION_MS,
+      sampleRate: 16000,
+      numberOfChannels: 1,
+      encodeBitRate: 48000,
+      format: 'mp3',
+      audioSource: 'auto'
+    })
+  },
+
+  stopRecording() {
+    if (!this.data.recording) return
+    finishRequested = true
+    this.clearRecordingTimer()
+    this.setData({
+      recording: false,
+      recordingPaused: false,
+      voiceBusy: true,
+      voiceStatus: '正在整理录音'
+    })
+    recorderManager.stop()
+  },
+
+  async handleRecorderStop(result: WechatMiniprogram.OnStopListenerResult) {
+    const lessonId = currentLessonId
+    if (!lessonId) return
+    const durationMs = Math.max(1, result.duration || Date.now() - segmentStartedAt)
+    const upload = this.queueSegmentUpload({
+      lessonId,
+      sequence: currentSegmentSequence,
+      durationMs,
+      filePath: result.tempFilePath
+    })
+    pendingUploads.push(upload)
+
+    if (!finishRequested) {
+      this.startNextRecordingSegment()
+      return
+    }
+    await this.finalizeCurrentLesson()
+  },
+
+  async queueSegmentUpload(segment: PendingVoiceUpload): Promise<void> {
+    try {
+      await this.uploadWithRetry(segment)
+    } catch (error) {
+      console.warn('Voice segment upload failed', error)
+      const saved = await this.persistRecordingFile(segment)
+      failedLocalUploads.push(saved)
+      wx.setStorageSync(PENDING_UPLOADS_KEY, failedLocalUploads)
+      throw error
+    }
+  },
+
+  async uploadWithRetry(segment: PendingVoiceUpload): Promise<void> {
+    let lastError: unknown
+    for (let attempt = 0; attempt < 2; attempt += 1) {
+      try {
+        await uploadVoiceSegment(
+          segment.lessonId,
+          segment.filePath,
+          segment.sequence,
+          segment.durationMs
+        )
+        return
+      } catch (error) {
+        lastError = error
+        if (attempt === 0) await wait(800)
+      }
+    }
+    throw lastError
+  },
+
+  persistRecordingFile(segment: PendingVoiceUpload): Promise<PendingVoiceUpload> {
+    return new Promise((resolve) => {
+      wx.getFileSystemManager().saveFile({
+        tempFilePath: segment.filePath,
+        success: (result) => resolve({ ...segment, filePath: result.savedFilePath }),
+        fail: () => resolve(segment)
+      })
+    })
+  },
+
+  async retryPendingUploads() {
+    if (failedLocalUploads.length === 0) return
+    this.setData({ voiceBusy: true, voiceStatus: '正在重试上传' })
+    const remaining: PendingVoiceUpload[] = []
+    for (const segment of failedLocalUploads) {
+      try {
+        await this.uploadWithRetry(segment)
+        this.removeSavedFile(segment.filePath)
+      } catch {
+        remaining.push(segment)
+      }
+    }
+    failedLocalUploads = remaining
+    wx.setStorageSync(PENDING_UPLOADS_KEY, failedLocalUploads)
+    if (remaining.length > 0) {
+      this.setData({ voiceBusy: false, voiceStatus: '上传失败，点击重试', voiceActionable: true })
+      wx.showToast({ title: '录音上传失败，请检查网络', icon: 'none' })
+      return
+    }
+    await this.finalizeCurrentLesson()
+  },
+
+  removeSavedFile(filePath: string) {
+    wx.getFileSystemManager().unlink({ filePath, fail: () => undefined })
+  },
+
+  async finalizeCurrentLesson() {
+    const lessonId = currentLessonId
+    if (!lessonId) {
+      this.setData({ voiceBusy: false, recording: false })
+      return
+    }
+    const results = await Promise.allSettled(pendingUploads)
+    pendingUploads = []
+    if (results.some((result) => result.status === 'rejected') || failedLocalUploads.length > 0) {
+      this.setData({
+        recording: false,
+        voiceBusy: false,
+        voiceStatus: '上传失败，点击重试',
+        voiceActionable: true
+      })
+      return
+    }
+
+    try {
+      const finished = await finishVoiceLesson(lessonId)
+      await this.applyFinishedLesson(finished)
+      currentLessonId = null
+      currentSegmentSequence = 0
+      finishRequested = false
+      await this.loadActiveSession()
+    } catch (error) {
+      wx.showToast({ title: getApiErrorMessage(error), icon: 'none' })
+    } finally {
+      this.clearRecordingTimer()
+      this.setData({ recording: false, recordingPaused: false, voiceBusy: false, recordingTime: '00:00' })
+      this.syncVoicePresentation(this.data.session)
+    }
+  },
+
+  async applyFinishedLesson(finished: VoiceLessonFinished) {
+    if (finished.status === 'processing') {
+      wx.showToast({ title: '录音已上传，正在后台转写', icon: 'none' })
+      return
+    }
+    if (finished.status === 'failed') {
+      wx.showToast({ title: '部分录音转录失败，点击状态可重试', icon: 'none' })
+      return
+    }
+    if (!finished.canApplyDirectly) {
+      wx.showToast({ title: '录音已就绪，可生成反馈', icon: 'none' })
+      return
+    }
+
+    const merged = appendTranscript(this.data.draft.content, finished.transcript)
+    this.setData({ 'draft.content': merged.content, contentLength: merged.content.length })
+    if (!merged.truncated) {
+      await markVoiceLessonApplied(finished.id, merged.content)
+      wx.showToast({ title: '语音已转为文字', icon: 'none' })
+    } else {
+      wx.showToast({ title: '内容已达上限，请生成汇总反馈', icon: 'none' })
+    }
+  },
+
+  startRecordingTimer() {
+    this.clearRecordingTimer()
+    recordingTimer = setInterval(() => {
+      this.setData({ recordingTime: formatRecordingTime(Date.now() - lessonStartedAt) })
+    }, 1000) as unknown as number
+  },
+
+  clearRecordingTimer() {
+    if (recordingTimer === null) return
+    clearInterval(recordingTimer)
+    recordingTimer = null
+  },
+
+  async openVoiceDetails() {
+    if (this.data.recording || this.data.voiceBusy) return
+    if (failedLocalUploads.length > 0) {
+      await this.retryPendingUploads()
+      return
+    }
+    const session = this.data.session
+    if (!session || session.lessonCount === 0) return
+    const lines = session.lessons.map((lesson) => {
+      const status = lesson.status === 'ready' ? '已就绪' : lesson.status === 'failed' ? '转录失败' : '处理中'
+      return `第${lesson.sequence}节 · ${formatVoiceDuration(lesson.durationMs)} · ${status}`
+    })
+    const hasFailures = session.failedSegmentCount > 0
+    wx.showModal({
+      title: '语音记录',
+      content: lines.join('\n'),
+      showCancel: hasFailures,
+      cancelText: '关闭',
+      confirmText: hasFailures ? '重试失败项' : '知道了',
+      success: (result) => {
+        if (hasFailures && result.confirm) void this.retryFailedSegments()
+      }
+    })
+  },
+
+  async retryFailedSegments() {
+    const session = this.data.session
+    if (!session) return
+    const failedSegments = session.lessons.flatMap((lesson) =>
+      lesson.segments.filter((segment) => segment.status === 'failed')
+    )
+    if (failedSegments.length === 0) return
+    this.setData({ voiceBusy: true, voiceStatus: '正在重试转录' })
+    try {
+      for (const segment of failedSegments) await retryVoiceSegment(segment.id)
+      const failedLessonIds = session.lessons
+        .filter((lesson) => lesson.segments.some((segment) => segment.status === 'failed'))
+        .map((lesson) => lesson.id)
+      for (const lessonId of failedLessonIds) {
+        const finished = await finishVoiceLesson(lessonId)
+        await this.applyFinishedLesson(finished)
+      }
+      await this.loadActiveSession()
+    } catch (error) {
+      wx.showToast({ title: getApiErrorMessage(error), icon: 'none' })
+    } finally {
+      this.setData({ voiceBusy: false })
+      this.syncVoicePresentation(this.data.session)
+    }
   },
 
   async saveFeedback() {
+    this.clearDraftSaveTimer()
+    const session = this.data.session
+    if (session && session.processingSegmentCount > 0) {
+      wx.showToast({ title: '录音仍在转写，请稍候', icon: 'none' })
+      this.scheduleVoicePolling(session)
+      return
+    }
+    if (session?.needsGeneration) {
+      await this.generateFeedback()
+      return
+    }
+
     const content = this.data.draft.content.trim()
     if (!content) {
       wx.showToast({ title: '请填写反馈内容', icon: 'none' })
@@ -91,8 +694,20 @@ Page({
 
     this.setData({ saving: true })
     try {
-      await createFeedbackRecord({ ...this.data.draft, content })
-      this.setData({ draft: buildDraft(), selectedProfileIndex: 0, contentLength: 0 })
+      await createFeedbackRecord({
+        ...this.data.draft,
+        content,
+        feedbackSessionId: session?.id || null
+      })
+      this.setData({
+        draft: buildDraft(),
+        session: null,
+        selectedProfileIndex: 0,
+        contentLength: 0,
+        voiceStatus: '语音录入',
+        voiceActionable: false,
+        primaryActionText: '保存反馈'
+      })
       wx.showToast({ title: '反馈已保存', icon: 'success' })
       setTimeout(() => wx.switchTab({ url: '/pages/records/index' }), 450)
     } catch (error) {
@@ -102,6 +717,26 @@ Page({
     }
   },
 
+  async generateFeedback() {
+    const session = this.data.session
+    if (!session) return
+    this.setData({ generating: true, voiceBusy: true, voiceStatus: '正在生成反馈' })
+    try {
+      const result = await generateFeedbackFromVoice(session.id, this.data.draft.content)
+      this.setData({
+        'draft.content': result.content,
+        contentLength: result.content.length
+      })
+      await this.loadActiveSession()
+      wx.showToast({ title: '反馈已生成，可继续修改', icon: 'none' })
+    } catch (error) {
+      wx.showToast({ title: getApiErrorMessage(error), icon: 'none' })
+    } finally {
+      this.setData({ generating: false, voiceBusy: false })
+      this.syncVoicePresentation(this.data.session)
+    }
+  },
+
   goStudents() {
     wx.switchTab({ url: '/pages/students/index' })
   }

pages/feedback/index.wxml

@@ -12,12 +12,12 @@
   </view>
 
   <view class="feedback-form surface">
-    <picker class="feedback-field" mode="selector" range="{{profileOptions}}" value="{{selectedProfileIndex}}" disabled="{{loading}}" bindchange="onProfileChange">
+    <picker class="feedback-field" mode="selector" range="{{profileOptions}}" value="{{selectedProfileIndex}}" disabled="{{loading || recording || voiceBusy}}" bindchange="onProfileChange">
       <text class="form-label">学生档案</text>
       <text class="form-value">{{loading ? '正在加载...' : profileOptions[selectedProfileIndex]}}</text>
     </picker>
 
-    <picker class="feedback-field" mode="date" value="{{draft.feedbackDate}}" bindchange="onDateChange">
+    <picker class="feedback-field" mode="date" value="{{draft.feedbackDate}}" disabled="{{recording || voiceBusy}}" bindchange="onDateChange">
       <text class="form-label">反馈日期</text>
       <text class="form-value">{{draft.feedbackDate}}</text>
     </picker>
@@ -31,12 +31,30 @@
         class="feedback-textarea"
         value="{{draft.content}}"
         maxlength="2000"
+        disabled="{{recording || voiceBusy}}"
         placeholder="记录课堂完成情况、进步与后续建议"
         bindinput="onContentInput"
       />
+
+      <view class="voice-input-row">
+        <button
+          class="voice-button {{recording ? 'voice-button-recording' : ''}}"
+          disabled="{{saving || voiceBusy}}"
+          bindtap="toggleRecording"
+          aria-label="{{recording ? '结束录音' : '开始录音'}}"
+        >
+          <text class="voice-button-icon">{{recording ? '■' : '●'}}</text>
+          <text>{{recording ? '结束录音' : '语音录入'}}</text>
+        </button>
+        <view class="voice-state {{voiceActionable ? 'voice-state-actionable' : ''}}" bindtap="openVoiceDetails">
+          <text class="voice-status">{{voiceStatus}}</text>
+          <text wx:if="{{recording}}" class="recording-time">{{recordingTime}}</text>
+          <text wx:elif="{{voiceActionable}}" class="voice-state-arrow">›</text>
+        </view>
+      </view>
     </view>
 
-    <button class="save-feedback primary-button" loading="{{saving}}" disabled="{{saving}}" bindtap="saveFeedback">保存反馈</button>
+    <button class="save-feedback primary-button" loading="{{saving || generating}}" disabled="{{saving || generating || recording || voiceBusy || (session && session.processingSegmentCount > 0)}}" bindtap="saveFeedback">{{primaryActionText}}</button>
   </view>
 
   <view wx:if="{{!loading && profiles.length === 0}}" class="profiles-hint">

pages/feedback/index.wxss

@@ -42,6 +42,91 @@
   line-height: 1.6;
 }
 
+.voice-input-row {
+  display: flex;
+  align-items: center;
+  gap: 20rpx;
+  margin-top: 18rpx;
+  padding-top: 18rpx;
+  border-top: 2rpx solid #e7eef6;
+}
+
+.voice-button {
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  gap: 12rpx;
+  width: 220rpx;
+  height: 72rpx;
+  margin: 0;
+  padding: 0 20rpx;
+  border: 2rpx solid #b8cbe1;
+  border-radius: 12rpx;
+  background: #ffffff;
+  color: #2b67e8;
+  font-size: 26rpx;
+  font-weight: 600;
+  line-height: 1;
+}
+
+.voice-button::after {
+  border: 0;
+}
+
+.voice-button[disabled] {
+  background: #f4f7fa;
+  color: #99a5b3;
+  opacity: 1;
+}
+
+.voice-button-recording,
+.voice-button-recording[disabled] {
+  border-color: #e46b65;
+  background: #fff4f3;
+  color: #c83f39;
+}
+
+.voice-button-icon {
+  font-size: 20rpx;
+  line-height: 1;
+}
+
+.voice-state {
+  display: flex;
+  flex: 1;
+  align-items: center;
+  justify-content: space-between;
+  min-width: 0;
+  color: #65758a;
+  font-size: 24rpx;
+}
+
+.voice-status {
+  flex: 1;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+
+.voice-state-actionable {
+  min-height: 64rpx;
+}
+
+.voice-state-arrow {
+  flex: none;
+  margin-left: 10rpx;
+  color: #7f8da0;
+  font-size: 38rpx;
+  line-height: 1;
+}
+
+.recording-time {
+  flex: none;
+  margin-left: 12rpx;
+  color: #c83f39;
+  font-variant-numeric: tabular-nums;
+}
+
 .save-feedback {
   margin-top: 24rpx;
 }

pages/profile/index.ts

@@ -41,9 +41,21 @@ Page({
     this.setData({ testing: true, connectionText: '正在检测服务...', connectionTone: 'idle' })
     try {
       const health = await getHealth()
+      let connectionText = '服务、数据库和语音转录正常'
+      let connectionTone = 'success'
+      if (!health.databaseConfigured) {
+        connectionText = '服务正常，数据库尚未配置'
+        connectionTone = 'warning'
+      } else if (!health.speechConfigured) {
+        connectionText = '服务和数据库正常，语音转录尚未配置'
+        connectionTone = 'warning'
+      } else if (!health.speechAvailable) {
+        connectionText = '服务和数据库正常，本地语音模型尚未就绪'
+        connectionTone = 'warning'
+      }
       this.setData({
-        connectionText: health.databaseConfigured ? '服务和数据库连接正常' : '服务正常，数据库尚未配置',
-        connectionTone: health.databaseConfigured ? 'success' : 'warning'
+        connectionText,
+        connectionTone
       })
     } catch (error) {
       this.setData({ connectionText: getApiErrorMessage(error), connectionTone: 'error' })

project.config.json

@@ -37,6 +37,10 @@
         "value": "server",
         "type": "folder"
       },
+      {
+        "value": "asr-service",
+        "type": "folder"
+      },
       {
         "value": "node_modules",
         "type": "folder"
@@ -44,6 +48,14 @@
       {
         "value": "graphify-out",
         "type": "folder"
+      },
+      {
+        "value": ".git",
+        "type": "folder"
+      },
+      {
+        "value": "compose.asr.yml",
+        "type": "file"
       }
     ],
     "include": []
@@ -51,4 +63,4 @@
   "appid": "wx3fe11e262a4b4885",
   "simulatorPluginLibVersion": {},
   "editorSetting": {}
-}
+}

utils/api.ts

@@ -1,11 +1,18 @@
 import type {
   FeedbackDraft,
   FeedbackRecord,
+  FeedbackSession,
+  GeneratedFeedback,
   HealthStatus,
   StudentDraft,
   StudentProfile,
   StudentProfileDefaults,
-  Term
+  Term,
+  VoiceAudioSegment,
+  VoiceLesson,
+  VoiceLessonCreated,
+  VoiceLessonFinished,
+  VoiceProcessingStatus
 } from './types'
 
 const DEFAULT_API_BASE_URL = 'http://127.0.0.1:8080'
@@ -50,6 +57,66 @@ interface RawFeedbackRecord {
 interface RawHealthStatus {
   status: string
   database_configured: boolean
+  speech_configured: boolean
+  speech_available: boolean
+  speech_provider: string | null
+  summary_configured: boolean
+}
+
+interface RawVoiceAudioSegment {
+  id: string
+  sequence: number
+  duration_ms: number
+  status: 'processing' | 'ready' | 'failed'
+  error_message: string | null
+}
+
+interface RawVoiceLesson {
+  id: string
+  sequence: number
+  status: VoiceProcessingStatus
+  duration_ms: number
+  applied_directly: boolean
+  included_in_generation: boolean
+  segments: RawVoiceAudioSegment[]
+  started_at: string
+  ended_at: string | null
+}
+
+interface RawFeedbackSession {
+  id: string
+  profile_id: string | null
+  feedback_date: string
+  content: string
+  status: 'active' | 'finalized'
+  lesson_count: number
+  total_duration_ms: number
+  processing_segment_count: number
+  failed_segment_count: number
+  needs_generation: boolean
+  lessons: RawVoiceLesson[]
+  created_at: string
+  updated_at: string
+}
+
+interface RawVoiceLessonCreated {
+  id: string
+  sequence: number
+  status: VoiceProcessingStatus
+  started_at: string
+}
+
+interface RawVoiceLessonFinished {
+  id: string
+  status: VoiceProcessingStatus
+  duration_ms: number
+  transcript: string
+  can_apply_directly: boolean
+}
+
+interface RawGeneratedFeedback {
+  content: string
+  method: 'llm' | 'extractive'
 }
 
 type HttpMethod = 'GET' | 'POST' | 'PUT' | 'DELETE'
@@ -81,17 +148,17 @@ export function getApiErrorMessage(error: unknown): string {
   if (!(error instanceof ApiRequestError)) return '请求失败，请稍后重试'
   if (error.statusCode === 401) return '开发身份未通过后端校验'
   if (error.statusCode === 404) return '请求的数据不存在或无权访问'
-  if (error.statusCode === 503) return '后端尚未配置数据库连接'
+  if (error.statusCode === 503 && error.message.includes('database')) return '后端尚未配置数据库连接'
   return error.message
 }
 
-function request<T>(path: string, method: HttpMethod = 'GET', data?: object): Promise<T> {
+function request<T>(path: string, method: HttpMethod = 'GET', data?: object, timeout = 10000): Promise<T> {
   return new Promise((resolve, reject) => {
     wx.request({
       url: `${getApiBaseUrl()}${path}`,
       method,
       data,
-      timeout: 10000,
+      timeout,
       header: {
         'Content-Type': 'application/json',
         'X-User-Id': DEVELOPMENT_USER_ID
@@ -165,11 +232,57 @@ function mapFeedbackRecord(record: RawFeedbackRecord): FeedbackRecord {
   }
 }
 
+function mapVoiceSegment(segment: RawVoiceAudioSegment): VoiceAudioSegment {
+  return {
+    id: segment.id,
+    sequence: segment.sequence,
+    durationMs: segment.duration_ms,
+    status: segment.status,
+    errorMessage: segment.error_message
+  }
+}
+
+function mapVoiceLesson(lesson: RawVoiceLesson): VoiceLesson {
+  return {
+    id: lesson.id,
+    sequence: lesson.sequence,
+    status: lesson.status,
+    durationMs: lesson.duration_ms,
+    appliedDirectly: lesson.applied_directly,
+    includedInGeneration: lesson.included_in_generation,
+    segments: lesson.segments.map(mapVoiceSegment),
+    startedAt: parseTimestamp(lesson.started_at),
+    endedAt: lesson.ended_at ? parseTimestamp(lesson.ended_at) : null
+  }
+}
+
+function mapFeedbackSession(session: RawFeedbackSession): FeedbackSession {
+  return {
+    id: session.id,
+    profileId: session.profile_id,
+    feedbackDate: session.feedback_date,
+    content: session.content,
+    status: session.status,
+    lessonCount: session.lesson_count,
+    totalDurationMs: session.total_duration_ms,
+    processingSegmentCount: session.processing_segment_count,
+    failedSegmentCount: session.failed_segment_count,
+    needsGeneration: session.needs_generation,
+    lessons: session.lessons.map(mapVoiceLesson),
+    createdAt: parseTimestamp(session.created_at),
+    updatedAt: parseTimestamp(session.updated_at)
+  }
+}
+
 export async function getHealth(): Promise<HealthStatus> {
   const result = await request<RawHealthStatus>('/health')
   return {
     status: result.status,
-    databaseConfigured: result.database_configured
+    databaseConfigured: result.database_configured,
+    speechConfigured: result.speech_configured,
+    speechAvailable: result.speech_available,
+    speechProvider: result.speech_provider,
+    summaryConfigured: result.summary_configured
   }
 }
 
@@ -209,7 +322,8 @@ export async function createFeedbackRecord(draft: FeedbackDraft): Promise<Feedba
   const result = await request<RawFeedbackRecord>('/api/v1/feedback-records', 'POST', {
     profile_id: draft.profileId,
     feedback_date: draft.feedbackDate,
-    content: draft.content
+    content: draft.content,
+    feedback_session_id: draft.feedbackSessionId || null
   })
   return mapFeedbackRecord(result)
 }
@@ -217,3 +331,134 @@ export async function createFeedbackRecord(draft: FeedbackDraft): Promise<Feedba
 export async function deleteFeedbackRecord(recordId: string): Promise<void> {
   await request<unknown>(`/api/v1/feedback-records/${encodeURIComponent(recordId)}`, 'DELETE')
 }
+
+export async function getActiveFeedbackSession(profileId: string | null): Promise<FeedbackSession | null> {
+  const query = profileId ? `?profile_id=${encodeURIComponent(profileId)}` : ''
+  const result = await request<RawFeedbackSession | null>(`/api/v1/feedback-sessions/active${query}`)
+  return result ? mapFeedbackSession(result) : null
+}
+
+export async function ensureFeedbackSession(
+  profileId: string | null,
+  feedbackDate: string,
+  content: string
+): Promise<FeedbackSession> {
+  const result = await request<RawFeedbackSession>('/api/v1/feedback-sessions', 'POST', {
+    profile_id: profileId,
+    feedback_date: feedbackDate,
+    content
+  })
+  return mapFeedbackSession(result)
+}
+
+export async function updateFeedbackSession(
+  sessionId: string,
+  feedbackDate: string,
+  content: string
+): Promise<FeedbackSession> {
+  const result = await request<RawFeedbackSession>(
+    `/api/v1/feedback-sessions/${encodeURIComponent(sessionId)}`,
+    'PUT',
+    { feedback_date: feedbackDate, content }
+  )
+  return mapFeedbackSession(result)
+}
+
+export async function createVoiceLesson(sessionId: string): Promise<VoiceLessonCreated> {
+  const result = await request<RawVoiceLessonCreated>(
+    `/api/v1/feedback-sessions/${encodeURIComponent(sessionId)}/lessons`,
+    'POST'
+  )
+  return {
+    id: result.id,
+    sequence: result.sequence,
+    status: result.status,
+    startedAt: parseTimestamp(result.started_at)
+  }
+}
+
+export function uploadVoiceSegment(
+  lessonId: string,
+  filePath: string,
+  sequence: number,
+  durationMs: number
+): Promise<VoiceAudioSegment> {
+  return new Promise((resolve, reject) => {
+    wx.uploadFile({
+      url: `${getApiBaseUrl()}/api/v1/feedback-lessons/${encodeURIComponent(lessonId)}/segments`,
+      filePath,
+      name: 'audio',
+      formData: {
+        sequence: String(sequence),
+        duration_ms: String(Math.max(1, Math.round(durationMs))),
+        audio_format: 'mp3'
+      },
+      timeout: 120000,
+      header: { 'X-User-Id': DEVELOPMENT_USER_ID },
+      success(response) {
+        let body: RawVoiceAudioSegment | ApiErrorBody
+        try {
+          body = JSON.parse(response.data) as RawVoiceAudioSegment | ApiErrorBody
+        } catch {
+          reject(new ApiRequestError('服务返回了无效的录音响应', response.statusCode))
+          return
+        }
+        if (response.statusCode >= 200 && response.statusCode < 300) {
+          resolve(mapVoiceSegment(body as RawVoiceAudioSegment))
+          return
+        }
+        reject(new ApiRequestError((body as ApiErrorBody).error || `服务返回 HTTP ${response.statusCode}`, response.statusCode))
+      },
+      fail(error) {
+        reject(new ApiRequestError(`录音上传失败：${error.errMsg}`))
+      }
+    })
+  })
+}
+
+export async function finishVoiceLesson(lessonId: string): Promise<VoiceLessonFinished> {
+  const result = await request<RawVoiceLessonFinished>(
+    `/api/v1/feedback-lessons/${encodeURIComponent(lessonId)}/finish`,
+    'POST',
+    undefined,
+    120000
+  )
+  return {
+    id: result.id,
+    status: result.status,
+    durationMs: result.duration_ms,
+    transcript: result.transcript,
+    canApplyDirectly: result.can_apply_directly
+  }
+}
+
+export async function markVoiceLessonApplied(lessonId: string, content: string): Promise<void> {
+  await request<unknown>(
+    `/api/v1/feedback-lessons/${encodeURIComponent(lessonId)}/mark-applied`,
+    'POST',
+    { content }
+  )
+}
+
+export async function retryVoiceSegment(segmentId: string): Promise<VoiceAudioSegment> {
+  const result = await request<RawVoiceAudioSegment>(
+    `/api/v1/audio-segments/${encodeURIComponent(segmentId)}/retry`,
+    'POST',
+    undefined,
+    120000
+  )
+  return mapVoiceSegment(result)
+}
+
+export async function generateFeedbackFromVoice(
+  sessionId: string,
+  existingContent: string
+): Promise<GeneratedFeedback> {
+  const result = await request<RawGeneratedFeedback>(
+    `/api/v1/feedback-sessions/${encodeURIComponent(sessionId)}/generate`,
+    'POST',
+    { existing_content: existingContent },
+    120000
+  )
+  return result
+}

utils/types.ts

@@ -44,9 +44,72 @@ export interface FeedbackDraft {
   profileId: string | null
   feedbackDate: string
   content: string
+  feedbackSessionId?: string | null
 }
 
 export interface HealthStatus {
   status: string
   databaseConfigured: boolean
+  speechConfigured: boolean
+  speechAvailable: boolean
+  speechProvider: string | null
+  summaryConfigured: boolean
+}
+
+export type VoiceProcessingStatus = 'recording' | 'processing' | 'ready' | 'failed'
+
+export interface VoiceAudioSegment {
+  id: string
+  sequence: number
+  durationMs: number
+  status: 'processing' | 'ready' | 'failed'
+  errorMessage: string | null
+}
+
+export interface VoiceLesson {
+  id: string
+  sequence: number
+  status: VoiceProcessingStatus
+  durationMs: number
+  appliedDirectly: boolean
+  includedInGeneration: boolean
+  segments: VoiceAudioSegment[]
+  startedAt: number
+  endedAt: number | null
+}
+
+export interface FeedbackSession {
+  id: string
+  profileId: string | null
+  feedbackDate: string
+  content: string
+  status: 'active' | 'finalized'
+  lessonCount: number
+  totalDurationMs: number
+  processingSegmentCount: number
+  failedSegmentCount: number
+  needsGeneration: boolean
+  lessons: VoiceLesson[]
+  createdAt: number
+  updatedAt: number
+}
+
+export interface VoiceLessonCreated {
+  id: string
+  sequence: number
+  status: VoiceProcessingStatus
+  startedAt: number
+}
+
+export interface VoiceLessonFinished {
+  id: string
+  status: VoiceProcessingStatus
+  durationMs: number
+  transcript: string
+  canApplyDirectly: boolean
+}
+
+export interface GeneratedFeedback {
+  content: string
+  method: 'llm' | 'extractive'
 }