From: lexicon <lexicon@local>
Date: Fri, 21 Aug 2026 12:06:01 +0000 (+0800)
Subject: V1: TTS 切换 Edge 神经嗓音(Sonia)+缩写展开; 修复 off_ed 同步误删进度; 补回 8/20 丢失记录
X-Git-Url: http://acesimba.cloud/gitweb/?a=commitdiff_plain;h=b9d4077883dec8b0c588af5c84bc5de334d0ff6d;p=words.git

V1: TTS 切换 Edge 神经嗓音(Sonia)+缩写展开; 修复 off_ed 同步误删进度; 补回 8/20 丢失记录

- backend/main.py: /api/tts 引擎由有道/百度改为 Edge 神经嗓音(en-GB-SoniaNeural)首选,
  有道/百度作兜底; 合成前做无歧义缩写展开(sth->something 等); 缓存键加 edge_sonia_ 前缀
- frontend/public/sw.js: CACHE_VERSION v1->v2, 强制重新拉取 TTS 音频避免旧难听嗓音残留
- frontend/src/sync.js: 删除 phase 2b 破坏性删除逻辑, 同步改为只补不删(布尔并集),
  修复 Review2 等已完成进度被静默丢弃的问题
- backend/db/bcd.db: 修正 8/20 云端记录(G54 review3 补回; G56 review2/G57 review1 撤销)
---

diff --git a/.gitignore b/.gitignore
index aabb8b3..8b60d53 100644
--- a/.gitignore
+++ b/.gitignore
@@ -20,3 +20,7 @@ backend/db/tts_cache/
 
 # certbot 一次性 DNS 挑战值（临时文件，不入库）
 deploy/dns-txt-value.txt
+
+# 导出/打包产物（供下载，不入库）
+words_package.zip
+words_export.xlsx
diff --git a/backend/db/bcd.db b/backend/db/bcd.db
index b54b828..ed30f86 100644
Binary files a/backend/db/bcd.db and b/backend/db/bcd.db differ
diff --git a/backend/main.py b/backend/main.py
index 4646339..fd91b1f 100644
--- a/backend/main.py
+++ b/backend/main.py
@@ -313,18 +313,51 @@ async def get_picture(pic_id: int):
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))
 
-# --- TTS 代理（有道美式发音 + 本地缓存） ---
+# --- TTS 代理（Edge 神经嗓音首选 + 有道/百度兜底 + 缩写展开） ---
 TTS_CACHE_DIR = os.path.join(BASE_DIR, "db", "tts_cache")
 os.makedirs(TTS_CACHE_DIR, exist_ok=True)
 
-def _build_tts_response(word: str):
+# 选用嗓音：英式 RP（接近高考听力风格，免费、无需 key）
+TTS_VOICE = "en-GB-SoniaNeural"
+# 缓存文件名前缀：换嗓音/换引擎时旧缓存不会被误用（避免仍播放旧嗓音）
+TTS_CACHE_PREFIX = "edge_sonia_"
+
+# 常见无歧义缩写 -> 完整读音（合成前展开，避免把 sth 拼读成三个字母）
+_ABBREV = {
+    "sth": "something", "sb": "somebody", "sbdy": "somebody",
+    "esp": "especially", "etc": "et cetera", "eg": "for example",
+    "ie": "that is", "approx": "approximately", "info": "information",
+    "num": "number", "max": "maximum", "min": "minimum",
+    "ext": "extension", "dept": "department", "gov": "government",
+    "lab": "laboratory", "exam": "examination", "gym": "gymnasium",
+    "vs": "versus", "pref": "prefix", "suff": "suffix", "adb": "adverb",
+}
+import re as _re
+def _expand_abbrev(text: str) -> str:
+    def _rep(m):
+        w = m.group(0)
+        return _ABBREV.get(w.lower(), w)
+    # 仅整词匹配替换，避免误改正常单词（如 clothes 不会被拆）
+    return _re.sub(r"[A-Za-z]+", _rep, text)
+
+async def _gen_edge_tts(text: str, voice: str) -> bytes:
+    import io, edge_tts
+    buf = io.BytesIO()
+    communicate = edge_tts.Communicate(text, voice)
+    async for chunk in communicate.stream():
+        if chunk.get("type") == "audio":
+            buf.write(chunk["data"])
+    return buf.getvalue()
+
+async def _build_tts_response(word: str):
     word = word.strip()
     if not word:
         raise HTTPException(status_code=400, detail="empty word")
 
-    # 缓存文件名：小写 + 下划线，避免特殊字符
-    safe_name = "".join(c if c.isalnum() or c in "-_" else "_" for c in word.lower())
-    cache_path = os.path.join(TTS_CACHE_DIR, f"{safe_name}.mp3")
+    # 缩写展开后再合成（如 sth -> something），缓存也按展开后的文本
+    spoken = _expand_abbrev(word)
+    safe_name = "".join(c if c.isalnum() or c in "-_" else "_" for c in spoken.lower())
+    cache_path = os.path.join(TTS_CACHE_DIR, f"{TTS_CACHE_PREFIX}{safe_name}.mp3")
 
     # 命中缓存直接返回
     if os.path.exists(cache_path):
@@ -332,29 +365,33 @@ def _build_tts_response(word: str):
             data = f.read()
         return Response(content=data, media_type="audio/mpeg")
 
-    # 多源 TTS：先有道（type=2 美式），失败回退百度翻译（本机可达，可覆盖有道无法合成的短语，无需 key）
-    sources = [
-        ("youdao", "https://dict.youdao.com/dictvoice?audio=" + urllib.parse.quote(word) + "&type=2",
-         {"User-Agent": "Mozilla/5.0"}),
-        ("baidu", "https://fanyi.baidu.com/gettts?lan=en&text=" + urllib.parse.quote(word) + "&spd=3",
-         {"User-Agent": "Mozilla/5.0", "Referer": "https://fanyi.baidu.com/"}),
-    ]
     data = None
     last_err = None
-    for name, url, headers in sources:
+    # 1) Edge 神经嗓音（首选）
+    try:
+        data = await _gen_edge_tts(spoken, TTS_VOICE)
+    except Exception as e:
+        last_err = f"edge: {e}"
+    # 2) 兜底: 有道（美式）
+    if not data:
         try:
-            req = urllib.request.Request(url, headers=headers)
+            url = "https://dict.youdao.com/dictvoice?audio=" + urllib.parse.quote(spoken) + "&type=2"
+            req = urllib.request.Request(url, headers={"User-Agent": "Mozilla/5.0"})
             with urllib.request.urlopen(req, timeout=10) as resp:
-                ct = resp.headers.get("Content-Type", "")
-                if ct.startswith("application/json"):
-                    last_err = f"{name} returned json (no audio)"
-                    continue
-                data = resp.read()
-            if data:
-                break
-        except urllib.error.URLError as e:
-            last_err = f"{name}: {e}"
-            continue
+                if not resp.headers.get("Content-Type", "").startswith("application/json"):
+                    data = resp.read()
+        except Exception as e:
+            last_err = f"youdao: {e}"
+    # 3) 兜底: 百度
+    if not data:
+        try:
+            url = "https://fanyi.baidu.com/gettts?lan=en&text=" + urllib.parse.quote(spoken) + "&spd=3"
+            req = urllib.request.Request(url, headers={"User-Agent": "Mozilla/5.0", "Referer": "https://fanyi.baidu.com/"})
+            with urllib.request.urlopen(req, timeout=10) as resp:
+                if not resp.headers.get("Content-Type", "").startswith("application/json"):
+                    data = resp.read()
+        except Exception as e:
+            last_err = f"baidu: {e}"
 
     if not data:
         raise HTTPException(status_code=502, detail=f"TTS upstream error: {last_err}")
@@ -371,13 +408,13 @@ def _build_tts_response(word: str):
 
 @app.get("/api/tts/{word}")
 async def get_tts(word: str):
-    return _build_tts_response(word)
+    return await _build_tts_response(word)
 
 
 @app.get("/api/tts")
 async def get_tts_query(w: str = ""):
     # 含 / 的短语（如 "in the/a form of"）走查询参数，避免与 URL 路径分隔符冲突
-    return _build_tts_response(w)
+    return await _build_tts_response(w)
 
 @app.post("/api/progress/update")
 async def update_progress(data: ProgressUpdate):
diff --git a/frontend/public/sw.js b/frontend/public/sw.js
index beef00e..94ba4c5 100644
--- a/frontend/public/sw.js
+++ b/frontend/public/sw.js
@@ -5,7 +5,7 @@
 //   - navigation (HTML)         → network-first, 失败回退 cached index.html
 //   - 静态资源 (JS/CSS/img/font) → cache-first
 
-const CACHE_VERSION = 'words-pwa-v1';
+const CACHE_VERSION = 'words-pwa-v2';
 
 self.addEventListener('install', (event) => {
     event.waitUntil(
diff --git a/frontend/src/sync.js b/frontend/src/sync.js
index 7009d23..99df80e 100644
--- a/frontend/src/sync.js
+++ b/frontend/src/sync.js
@@ -6,7 +6,7 @@
 
 import { api } from './api';
 import {
-    getDirtyProgress, clearDirty, getAllProgress, putProgress, deleteProgress,
+    getDirtyProgress, clearDirty, getAllProgress, putProgress,
     putWords, setMeta, getMeta, setLearnableGroups, getAllWordsFromIDB, getWordsCount,
 } from './db';
 
@@ -79,17 +79,12 @@ export async function syncNow() {
         }
     }
 
-    // 阶段2b: 服务端已删除(且本地已确认/非 dirty)的进度, 本地同步删除。
-    // 仅在成功拉取到非空集合时才执行, 避免服务端异常返回空时误删本地进度;
-    // 仅删非 dirty 记录(本地未推送完成的进度绝不删, 保护离线工作)。
-    if (serverEvents.length > 0) {
-        const serverKeys = new Set(serverEvents.map(e => `${e.group_id}:${e.op_type}`));
-        for (const p of localAll) {
-            if (!p.dirty && !serverKeys.has(p.key)) {
-                await deleteProgress(p.group_id, p.op_type);
-            }
-        }
-    }
+    // 阶段2b: 不再自动删除本地进度。
+    // 原逻辑会把"服务端 pull 中不存在的本地非 dirty 记录"删掉，用于"撤销"回写；
+    // 但布尔并集语义下，服务端 pull 缺失≠被撤销(可能是推送时序/服务端异常)，
+    // 这会静默丢掉已完成的进度(如 Review2)。回归离线方案原始设计「不删本地」，
+    // 同步只做布尔并集(只补不删)，保证任何已完成的 (group,op) 永不丢失。
+    // 如需撤销，应在 /api/sync/pull 显式返回"已删除清单"再据此删除，而非按缺失推断。
 
     // 阶段3: 首次 warmup 词库 (仅 IDB 空时, 避免每次重拉 1852 词)
     if ((await getWordsCount()) === 0) {