weflow-cli 1.16.0 → 1.17.0

package/scripts/voice_transcribe.py

@@ -1,139 +1,139 @@
-"""批量转写语音消息，产出一份可检索的清单（`output/.voice_index/voices.jsonl`）。
-
-为什么要有它：`wechat_voice.py` 只能一条会话一条会话地转，转完就躺在一个按**内容哈希**
-命名的缓存里——**知道"这段音频说了什么"，但不知道"是哪条消息、谁在什么时候发的"**。
-而检索需要的正是后者。
-
-所以这里做三件事：
-  1. 从 `media_*.db` 的 `VoiceInfo` 按时间取语音（带上会话、发送者、时间）；
-  2. 调 `wechat_voice.ensure_transcripts` 填空缺（断点续传：已转过的按内容命中缓存）；
-  3. 把 `(消息, 时间, 会话, 文本)` 落成 JSONL，供 `phrase-build` 并进 FTS。
-
-用法：
-    python scripts/voice_transcribe.py --days 30
-    python scripts/voice_transcribe.py --days 7 --limit 5      # 先试几条
-    python scripts/voice_transcribe.py --days 30 --model tiny  # 更快、更糙
-"""
-from __future__ import annotations
-
-import argparse
-import json
-import pathlib
-import sys
-import time
-from datetime import datetime, timezone, timedelta
-
-SCRIPT_DIR = pathlib.Path(__file__).resolve().parent
-sys.path.insert(0, str(SCRIPT_DIR))
-
-from _utils import load_config, decrypt_lock  # noqa: E402
-import wechat_voice as wv  # noqa: E402
-
-ROOT = SCRIPT_DIR.parent
-OUT_DIR = ROOT / 'output' / '.voice_index'
-CACHE_DIR = ROOT / 'output' / '.voice_cache'
-TZ = timezone(timedelta(hours=8))
-
-
-def _keys(cfg):
-    """配置里的密钥是 `lock:` 密文，必须先解密 —— 直接拿密文去派生只会得到"不是数据库"。"""
-    key = decrypt_lock(cfg.get('ntKey', '') or '')
-    salt = decrypt_lock(cfg.get('ntSalt', '') or '')
-    pw = decrypt_lock(cfg.get('favPassphrase', '') or cfg.get('decryptKey', '') or '')
-    return key, salt, pw
-
-
-def collect(cfg, days, limit=0):
-    """[(shard, chat_name_id, local_id, create_time, blob, name)]，按时间倒序。"""
-    key, salt, pw = _keys(cfg)
-    db = cfg.get('ntDbPath') or ''
-    if not db:
-        raise SystemExit('配置里没有 ntDbPath')
-    since = int(time.time()) - days * 86400
-    rows = []
-    for shard in wv.voice_shards(db):
-        conn = None
-        try:
-            conn, cur = wv._open(shard, key, salt, pw)
-            cur.execute('SELECT rowid, user_name FROM Name2Id')
-            names = {r[0]: r[1] for r in cur.fetchall()}
-            cur.execute(
-                'SELECT chat_name_id, local_id, create_time, voice_data FROM VoiceInfo '
-                'WHERE voice_data IS NOT NULL AND create_time >= ? '
-                'ORDER BY create_time DESC', (since,))
-            for cid, lid, ts, blob in cur.fetchall():
-                rows.append((str(shard), cid, lid, int(ts or 0), blob,
-                         names.get(cid, str(cid))))
-        except Exception as exc:
-            print(f'[WARN] {pathlib.Path(shard).name} 读不了（{type(exc).__name__}: {exc}）',
-                  file=sys.stderr)
-        finally:
-            if conn:
-                conn.close()
-    rows.sort(key=lambda r: r[3], reverse=True)
-    if limit and limit > 0:
-        rows = rows[:limit]
-
-    # 会话名从 **contact.db** 补一层：消息库的 `Name2Id` 只有 id，语音清单里
-    # 只写 id 的话，搜索结果显示成 `wxid_xxx`（2026-10-22 第一版就是这样）。
-    try:
-        from network import load_graph
-        people, _rooms, _members = load_graph(cfg)
-        by_username = {p['username']: (p['remark'] or p['nick'] or p['username'])
-                       for p in people.values()}
-        rows = [(s, c, l, t, b, by_username.get(name, name)) for s, c, l, t, b, name in rows]
-    except Exception as exc:
-        print(f'[WARN] 读联系人失败（{type(exc).__name__}: {exc}），语音清单里显示 id',
-              file=sys.stderr)
-    return rows
-
-
-def main():
-    ap = argparse.ArgumentParser(description='批量转写微信语音，产出可检索清单')
-    ap.add_argument('--days', type=int, default=30, help='只转最近 N 天（默认 30）')
-    ap.add_argument('--limit', type=int, default=0, help='最多转写几条（0 = 不限）')
-    ap.add_argument('--model', default='small', help='faster-whisper 模型名或本地目录')
-    ap.add_argument('--language', default='zh', help='语言提示')
-    ap.add_argument('--cache-dir', default='', help='转写缓存目录（换模型时换一个，否则会命中旧模型的结果）')
-    args = ap.parse_args()
-
-    cfg = load_config()
-    rows = collect(cfg, args.days, args.limit)
-    print(f'最近 {args.days} 天语音: {len(rows)} 条')
-    if not rows:
-        return 0
-    cache_dir = args.cache_dir or str(CACHE_DIR)
-    pathlib.Path(cache_dir).mkdir(parents=True, exist_ok=True)
-
-    voices = {(r[5], r[2]): r[4] for r in rows}
-    started = time.time()
-    processed, cached, failed = wv.ensure_transcripts(
-        voices, cache_dir, model_name=args.model, language=args.language)
-    elapsed = time.time() - started
-    print(f'转写完成: 新增 {processed}，缓存命中 {cached}，失败 {failed}，耗时 {elapsed/60:.1f} 分钟')
-
-    cache = wv.TranscriptCache(cache_dir)
-    OUT_DIR.mkdir(parents=True, exist_ok=True)
-    out = OUT_DIR / 'voices.jsonl'
-    written = empty = 0
-    with out.open('w', encoding='utf-8') as fh:
-        for shard, cid, lid, ts, blob, name in rows:
-            text = (cache.get(wv.voice_key(blob)) or '').strip()
-            if not text:
-                empty += 1
-                continue
-            dt = datetime.fromtimestamp(ts, tz=TZ)
-            fh.write(json.dumps({
-                'id': f'voice:{name}:{lid}',
-                'talker': name,
-                'time': dt.strftime('%Y-%m-%d %H:%M'),
-                'ts': ts,
-                'text': text[:500],
-            }, ensure_ascii=False) + '\n')
-            written += 1
-    print(f'写入 {out}：{written} 条有文字（空转写/解码失败 {empty} 条）')
-    return 0
-
-
-if __name__ == '__main__':
-    sys.exit(main())
+"""批量转写语音消息，产出一份可检索的清单（`output/.voice_index/voices.jsonl`）。
+
+为什么要有它：`wechat_voice.py` 只能一条会话一条会话地转，转完就躺在一个按**内容哈希**
+命名的缓存里——**知道"这段音频说了什么"，但不知道"是哪条消息、谁在什么时候发的"**。
+而检索需要的正是后者。
+
+所以这里做三件事：
+  1. 从 `media_*.db` 的 `VoiceInfo` 按时间取语音（带上会话、发送者、时间）；
+  2. 调 `wechat_voice.ensure_transcripts` 填空缺（断点续传：已转过的按内容命中缓存）；
+  3. 把 `(消息, 时间, 会话, 文本)` 落成 JSONL，供 `phrase-build` 并进 FTS。
+
+用法：
+    python scripts/voice_transcribe.py --days 30
+    python scripts/voice_transcribe.py --days 7 --limit 5      # 先试几条
+    python scripts/voice_transcribe.py --days 30 --model tiny  # 更快、更糙
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import pathlib
+import sys
+import time
+from datetime import datetime, timezone, timedelta
+
+SCRIPT_DIR = pathlib.Path(__file__).resolve().parent
+sys.path.insert(0, str(SCRIPT_DIR))
+
+from _utils import load_config, decrypt_lock  # noqa: E402
+import wechat_voice as wv  # noqa: E402
+
+ROOT = SCRIPT_DIR.parent
+OUT_DIR = ROOT / 'output' / '.voice_index'
+CACHE_DIR = ROOT / 'output' / '.voice_cache'
+TZ = timezone(timedelta(hours=8))
+
+
+def _keys(cfg):
+    """配置里的密钥是 `lock:` 密文，必须先解密 —— 直接拿密文去派生只会得到"不是数据库"。"""
+    key = decrypt_lock(cfg.get('ntKey', '') or '')
+    salt = decrypt_lock(cfg.get('ntSalt', '') or '')
+    pw = decrypt_lock(cfg.get('favPassphrase', '') or cfg.get('decryptKey', '') or '')
+    return key, salt, pw
+
+
+def collect(cfg, days, limit=0):
+    """[(shard, chat_name_id, local_id, create_time, blob, name)]，按时间倒序。"""
+    key, salt, pw = _keys(cfg)
+    db = cfg.get('ntDbPath') or ''
+    if not db:
+        raise SystemExit('配置里没有 ntDbPath')
+    since = int(time.time()) - days * 86400
+    rows = []
+    for shard in wv.voice_shards(db):
+        conn = None
+        try:
+            conn, cur = wv._open(shard, key, salt, pw)
+            cur.execute('SELECT rowid, user_name FROM Name2Id')
+            names = {r[0]: r[1] for r in cur.fetchall()}
+            cur.execute(
+                'SELECT chat_name_id, local_id, create_time, voice_data FROM VoiceInfo '
+                'WHERE voice_data IS NOT NULL AND create_time >= ? '
+                'ORDER BY create_time DESC', (since,))
+            for cid, lid, ts, blob in cur.fetchall():
+                rows.append((str(shard), cid, lid, int(ts or 0), blob,
+                         names.get(cid, str(cid))))
+        except Exception as exc:
+            print(f'[WARN] {pathlib.Path(shard).name} 读不了（{type(exc).__name__}: {exc}）',
+                  file=sys.stderr)
+        finally:
+            if conn:
+                conn.close()
+    rows.sort(key=lambda r: r[3], reverse=True)
+    if limit and limit > 0:
+        rows = rows[:limit]
+
+    # 会话名从 **contact.db** 补一层：消息库的 `Name2Id` 只有 id，语音清单里
+    # 只写 id 的话，搜索结果显示成 `wxid_xxx`（2026-10-22 第一版就是这样）。
+    try:
+        from network import load_graph
+        people, _rooms, _members = load_graph(cfg)
+        by_username = {p['username']: (p['remark'] or p['nick'] or p['username'])
+                       for p in people.values()}
+        rows = [(s, c, l, t, b, by_username.get(name, name)) for s, c, l, t, b, name in rows]
+    except Exception as exc:
+        print(f'[WARN] 读联系人失败（{type(exc).__name__}: {exc}），语音清单里显示 id',
+              file=sys.stderr)
+    return rows
+
+
+def main():
+    ap = argparse.ArgumentParser(description='批量转写微信语音，产出可检索清单')
+    ap.add_argument('--days', type=int, default=30, help='只转最近 N 天（默认 30）')
+    ap.add_argument('--limit', type=int, default=0, help='最多转写几条（0 = 不限）')
+    ap.add_argument('--model', default='small', help='faster-whisper 模型名或本地目录')
+    ap.add_argument('--language', default='zh', help='语言提示')
+    ap.add_argument('--cache-dir', default='', help='转写缓存目录（换模型时换一个，否则会命中旧模型的结果）')
+    args = ap.parse_args()
+
+    cfg = load_config()
+    rows = collect(cfg, args.days, args.limit)
+    print(f'最近 {args.days} 天语音: {len(rows)} 条')
+    if not rows:
+        return 0
+    cache_dir = args.cache_dir or str(CACHE_DIR)
+    pathlib.Path(cache_dir).mkdir(parents=True, exist_ok=True)
+
+    voices = {(r[5], r[2]): r[4] for r in rows}
+    started = time.time()
+    processed, cached, failed = wv.ensure_transcripts(
+        voices, cache_dir, model_name=args.model, language=args.language)
+    elapsed = time.time() - started
+    print(f'转写完成: 新增 {processed}，缓存命中 {cached}，失败 {failed}，耗时 {elapsed/60:.1f} 分钟')
+
+    cache = wv.TranscriptCache(cache_dir)
+    OUT_DIR.mkdir(parents=True, exist_ok=True)
+    out = OUT_DIR / 'voices.jsonl'
+    written = empty = 0
+    with out.open('w', encoding='utf-8') as fh:
+        for shard, cid, lid, ts, blob, name in rows:
+            text = (cache.get(wv.voice_key(blob)) or '').strip()
+            if not text:
+                empty += 1
+                continue
+            dt = datetime.fromtimestamp(ts, tz=TZ)
+            fh.write(json.dumps({
+                'id': f'voice:{name}:{lid}',
+                'talker': name,
+                'time': dt.strftime('%Y-%m-%d %H:%M'),
+                'ts': ts,
+                'text': text[:500],
+            }, ensure_ascii=False) + '\n')
+            written += 1
+    print(f'写入 {out}：{written} 条有文字（空转写/解码失败 {empty} 条）')
+    return 0
+
+
+if __name__ == '__main__':
+    sys.exit(main())