fix(db): create upload_chunks before ALTERs + bump Soniox v5, v1.3.2

Fresh installs crashed on upload with "no such column: status". In
_migrate_v2 the upload_chunks ALTERs ran before the CREATE TABLE, so on
a fresh DB the ALTER hit "no such table" (silently dropped — the loop
only ignores "duplicate column") and the table was created without
status/error_message/segments_json. Create the child tables first and
declare the three columns inline; keep the ALTERs to top up old DBs.

Also bump Soniox to v5 (stt-async-v5 for upload, stt-rt-v5 for realtime)
— same price, better accuracy + speaker separation, name-only change.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: luongnd

CHANGELOG.md

@@ -2,6 +2,14 @@
 
 Tất cả thay đổi đáng chú ý của Scribble được ghi tại đây.
 
+## [1.3.2] - 2026-06-22
+
+### Fixed
+- Sửa lỗi `no such column: status` khi upload file trên máy cài mới. Migration tạo bảng `upload_chunks` sai thứ tự (ALTER chạy trước CREATE) khiến bảng thiếu cột `status/error_message/segments_json` — nay tạo bảng trước, khai báo cột inline, ALTER chỉ top-up cho DB cũ.
+
+### Changed
+- Nâng Soniox lên model v5 (`stt-async-v5` cho upload, `stt-rt-v5` cho ghi âm realtime) — chất lượng nhận diện và tách người nói tốt hơn, giá không đổi.
+
 ## [1.3.1] - 2026-06-09
 
 ### Fixed

package.json

@@ -1,7 +1,7 @@
 {
   "name": "voicescribe",
   "private": true,
-  "version": "1.3.1",
+  "version": "1.3.2",
   "type": "module",
   "scripts": {
     "dev": "vite",

src-python/db.py

@@ -109,24 +109,15 @@ def _migrate_v2(self):
         Existing rows get default values; realtime flow unaffected.
         """
         conn = self._conn()
-        migrations = [
-            "ALTER TABLE meetings ADD COLUMN source_type TEXT DEFAULT 'realtime'",
-            "ALTER TABLE meetings ADD COLUMN file_hash TEXT DEFAULT NULL",
-            "ALTER TABLE meetings ADD COLUMN source_filename TEXT DEFAULT NULL",
-            "CREATE INDEX IF NOT EXISTS idx_meetings_file_hash ON meetings(file_hash)",
-            # v1.2.13: per-chunk failure tracking + denormalized failed count
-            # for fast list queries (avoids COUNT(*) per meeting).
-            "ALTER TABLE meetings ADD COLUMN failed_chunks_count INTEGER DEFAULT 0",
-            "ALTER TABLE upload_chunks ADD COLUMN status TEXT DEFAULT NULL",
-            "ALTER TABLE upload_chunks ADD COLUMN error_message TEXT DEFAULT NULL",
-            "ALTER TABLE upload_chunks ADD COLUMN segments_json TEXT DEFAULT NULL",
-        ]
-        for sql in migrations:
-            try:
-                conn.execute(sql)
-            except sqlite3.OperationalError as e:
-                if "duplicate column" not in str(e).lower():
-                    log.warning("Migration skipped (%s): %s", sql[:60], e)
+
+        # Create child tables FIRST — before the ALTER migrations below.
+        # ORDER MATTERS: the upload_chunks ALTERs target this table. On a fresh
+        # DB, an ALTER that ran before CREATE threw "no such table", which the
+        # loop swallowed (it only ignores "duplicate column"). That left
+        # upload_chunks without status/error_message/segments_json and broke
+        # every fresh install with "no such column: status" on upload. The
+        # three columns are also declared inline here so a fresh DB is complete
+        # even when the ALTERs are no-ops.
 
         # ── Per-chunk progress (upload pipeline resume) ──
         # text IS NULL until STT completes; embedding BLOB is the raw 512×f32
@@ -141,6 +132,9 @@ def _migrate_v2(self):
                 text TEXT DEFAULT NULL,
                 embedding BLOB DEFAULT NULL,
                 updated_at DATETIME DEFAULT CURRENT_TIMESTAMP,
+                status TEXT DEFAULT NULL,
+                error_message TEXT DEFAULT NULL,
+                segments_json TEXT DEFAULT NULL,
                 PRIMARY KEY (meeting_id, chunk_idx)
             )
             """
@@ -170,6 +164,29 @@ def _migrate_v2(self):
         conn.execute(
             "CREATE INDEX IF NOT EXISTS idx_attachments_meeting ON meeting_attachments(meeting_id)"
         )
+
+        # Column top-ups for DBs created by older versions. The tables above
+        # already exist by now, so an upload_chunks ALTER either adds the
+        # missing column (old DB) or is ignored as "duplicate column".
+        migrations = [
+            "ALTER TABLE meetings ADD COLUMN source_type TEXT DEFAULT 'realtime'",
+            "ALTER TABLE meetings ADD COLUMN file_hash TEXT DEFAULT NULL",
+            "ALTER TABLE meetings ADD COLUMN source_filename TEXT DEFAULT NULL",
+            "CREATE INDEX IF NOT EXISTS idx_meetings_file_hash ON meetings(file_hash)",
+            # v1.2.13: per-chunk failure tracking + denormalized failed count
+            # for fast list queries (avoids COUNT(*) per meeting).
+            "ALTER TABLE meetings ADD COLUMN failed_chunks_count INTEGER DEFAULT 0",
+            "ALTER TABLE upload_chunks ADD COLUMN status TEXT DEFAULT NULL",
+            "ALTER TABLE upload_chunks ADD COLUMN error_message TEXT DEFAULT NULL",
+            "ALTER TABLE upload_chunks ADD COLUMN segments_json TEXT DEFAULT NULL",
+        ]
+        for sql in migrations:
+            try:
+                conn.execute(sql)
+            except sqlite3.OperationalError as e:
+                if "duplicate column" not in str(e).lower():
+                    log.warning("Migration skipped (%s): %s", sql[:60], e)
+
         conn.commit()
 
     # ─── Upload chunks (resume support) ───

src-python/services/upload_pipeline.py

@@ -76,7 +76,7 @@ def _normalize_stt_provider(raw: str | None) -> str:
 # setting `upload_max_duration_hours` if a user has legit >24h needs.
 DEFAULT_MAX_DURATION_HOURS = 24
 
-# Soniox stt-async-v4 has an ~5h cap per submission. We split at 3.5h to
+# Soniox stt-async-v5 has an ~5h cap per submission. We split at 3.5h to
 # leave a 30% safety buffer for network slowness, peak-load queueing, and
 # Soniox-side variance. Files ≤ 3.5h take the single-shot path (preserves
 # globally-consistent speaker diarization across the whole file).
@@ -647,7 +647,7 @@ async def _run_soniox_pipeline(
     duration_sec: float, tmp_root: Path,
 ) -> list[dict]:
     """Soniox STT pipeline. Auto-splits files > 3.5h into chunks because
-    Soniox stt-async-v4 caps at ~5h per submission.
+    Soniox stt-async-v5 caps at ~5h per submission.
 
     Returns transcript_parts ready for ``db.update_meeting(transcript=...)``.
       - File ≤ 3.5h: single-shot — Soniox does globally-consistent diarization
@@ -1561,7 +1561,7 @@ async def _process_one(chunk: AudioChunk):
 
             # Dispatch to the configured provider. Nvidia uses streaming gRPC
             # (offline_recognize unavailable for vi/zh). Soniox uses the
-            # async file API (stt-async-v4) with auto-cleanup.
+            # async file API (stt-async-v5) with auto-cleanup.
             if stt_provider == "nvidia":
                 stt_task = asyncio.to_thread(
                     transcribe_nvidia_streaming, str(chunk.path), nvidia_key, riva_lang

src-python/stt.py

@@ -510,7 +510,7 @@ def transcribe_soniox_file_id(
     )
     client = SonioxClient(api_key=api_key)
     config = CreateTranscriptionConfig(
-        model="stt-async-v4",
+        model="stt-async-v5",
         language_hints=hints,
         enable_speaker_diarization=True,
         enable_language_identification=False,
@@ -916,7 +916,7 @@ class SonioxStreamingSTT:
     - receive_events() runs on the calling thread (results generator)
     """
 
-    # Soniox stt-rt-v4 has a server-side session duration cap of 300 min (5h,
+    # Soniox stt-rt-v5 has a server-side session duration cap of 300 min (5h,
     # fixed — see docs/stt/rt/limits-and-quotas). When the cap fires OR the
     # connection idle-closes, the server closes the WS and our event iterator
     # exits silently. We auto-reconnect up to this many times.
@@ -962,7 +962,7 @@ def _open_session(self):
         from soniox.types import RealtimeSTTConfig, TranslationConfig
 
         config = RealtimeSTTConfig(
-            model="stt-rt-v4",
+            model="stt-rt-v5",
             audio_format="pcm_s16le",
             sample_rate=16000,
             num_channels=1,

src-tauri/Cargo.lock

@@ -1,6 +1,6 @@
 [package]
 name = "voicescribe"
-version = "1.3.1"
+version = "1.3.2"
 description = "Scribble - Meeting Minutes"
 authors = ["you"]
 edition = "2021"

src-tauri/Cargo.toml

@@ -1,7 +1,7 @@
 {
   "$schema": "https://schema.tauri.app/config/2",
   "productName": "Scribble",
-  "version": "1.3.1",
+  "version": "1.3.2",
   "identifier": "com.scribble.app",
   "build": {
     "beforeDevCommand": "pnpm dev",