Skip to content

Commit 36a78be

Browse files
meizhong986meizhong986
committed
ahead of dev 1.8.13
1 parent a985674 commit 36a78be

1 file changed

Lines changed: 2 additions & 2 deletions

File tree

‎docs/ISSUE_TRACKER_v1.8.x.md‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -25,7 +25,7 @@
2525
| **v1.8.12 RELEASED** | — | **2026-04-30**. Tagged `v1.8.12`, merged `dev_v1.8.12`→main, pushed. GitHub Release published with `.exe` + wheel. **20 commits** including WhisperSeg VAD, tight defaults retune, engine-split presets, F5 regression fix, anime ellipsis filter. |
2626
| **v1.8.11 RELEASED** | — | 2026-04-20. Tagged, merged, pushed. Stable prior release. |
2727
| **v1.8.10.post3 RELEASED** | — | 2026-04-09. Stable (older release). |
28-
| **NEEDS RESPONSE** | **2** | **#304, #311.** **rev46 batch flipped 5 → 2**: #305 (CN→EN translation), #309 (mklink white screen), #312 (VAD dump_params), #307 (sanitization Q) all responded; #313 + #300 responded earlier in rev46. Remaining: #304 (GTX 1050 Ti compat), #311 (FireRedVAD feature request). |
28+
| **NEEDS RESPONSE** | **1** | **#304** only. **#311 responded 05-01** with v1.9 parking decision (FireRedVAD held until WhisperSeg field feedback lands, then benchmark on Netflix-GT before positioning). |
2929
| **NEEDS FOLLOW-UP** | **2** | **#231, #264.** **rev46 batch flipped 4 → 0**: #287, #294, #302, #271 follow-up all responded with v1.8.12 retest pings + WhisperSeg recommendation. Remaining: #231 (Kaggle notebook clarification), #264 (cache location verified docs). |
3030
| **AWAITING REPLY** | 20 | **−10 vs rev46**: 10 stale AWAITING REPLY threads closed in rev47 sweep. Remaining 20 are recent (≤10 days). |
3131
| **SHIPPED in v1.8.12** | **8** | **F5 best_of regression** (#294, #302 cluster — needs user retest), **anime ellipsis filter** (anime-whisper artifact lines), **silero-v3.1 fallback alignment** (#297 cluster — fixes the "silero-v4.0 random fail" teijiIshida found), **TEN max_speech schema fix**, **anime chunk_threshold plumbing**, **`whisperseg` in `--qwen-segmenter` choices**, **engine-split sensitivity retune** (aggressive `no_speech_threshold` 0.77→0.84 — addresses #294/#302/#287 root cause), **WhisperSeg ONNX VAD backend** added. |
@@ -128,7 +128,7 @@ See `installer/RELEASE_NOTES_v1.8.10.post3.md` for full details.
128128
|---|-------|----------|--------|-------|
129129
| **#313** | Build error on 1.8.12 (Linux `uv sync` deadlock) | parheliamm | `AWAITING REPLY` | **NEW 2026-04-30** (~3 hours after v1.8.12 release). Linux aarch64 / py3.12 install fails: `bs-roformer-infer requires requests>=2.31` but pytorch index pins `requests==2.28.1`. **Same root cause as #300**. **Responded 04-30** ([comment-4358567076](https://github.com/meizhong986/WhisperJAV/issues/313#issuecomment-4358567076)): `UV_INDEX_STRATEGY=unsafe-best-match python install.py` workaround + v1.8.13 fix promise. **Real fix candidate for v1.8.13**: pass `--index-strategy unsafe-best-match` from install.py automatically, or pin `requests>=2.31` higher in resolution priority. |
130130
| **#312** | VAD Parameters dump_params shows silero values for TEN | TinyRick1489 | `AWAITING REPLY` | **NEW 04-28.** Sharp investigation. **Responded 04-30** ([comment-4358678182](https://github.com/meizhong986/WhisperJAV/issues/312#issuecomment-4358678182)): explained metadata-vs-runtime distinction — legacy resolver always emits silero presets (because `LEGACY_PIPELINES["fidelity"]["vad"] = "silero-v3.1"` is hard-coded), but actual TEN backend at runtime gets correct TEN config via firewall in `whisper_pro_asr.py`. Runtime is correct, only dump is misleading. v1.8.13 fix planned to make dump reflect runtime config. |
131-
| **#311** | Feature Request: Add FireRedVAD as a new VAD backend | kylesskim-sys | `NEEDS RESPONSE` | **NEW 04-28.** Industrial-grade SOTA VAD released 03-2026 by FireRedTeam (HF: `FireRedTeam/FireRedVAD`). Cites FLEURS-VAD-102 benchmark: F1=97.57 vs Silero=95.95 vs TEN=95.19; **False Alarm Rate 2.69% vs 9.41%/15.47%**. Claims dramatic FA reduction is exactly what JAV scenes with BGM/breathing need. Also has Audio Event Detection (AED) feature for speech vs music separation. Output format similar to existing VADs → integration "relatively straightforward". Strong feature request — file as v1.9+ candidate; respond with thanks + benchmark interest + acknowledge VAD evaluator tool can compare it side-by-side with existing backends. Tracked under Feature Requests. |
131+
| **#311** | Feature Request: Add FireRedVAD as a new VAD backend | kylesskim-sys | `AWAITING REPLY` | **NEW 04-28.** FireRedVAD industrial-grade SOTA VAD (Apache-2.0, 2.2 MB DFSMN model, PyTorch, HF + ModelScope). FLEURS-VAD-102: F1=97.57 (vs Silero 95.95, TEN 95.19), False Alarm Rate **2.69% vs 9.41%/15.47%**. AED feature for speech/singing/music separation. **Decision 05-01**: park for v1.9 series after gathering WhisperSeg field feedback first — stacking two new segmenters in consecutive releases makes attribution noisy. **Responded 05-01** ([comment-4359972786](https://github.com/meizhong986/WhisperJAV/issues/311#issuecomment-4359972786)): thanks + v1.9 candidate + reasoning (wait for WhisperSeg feedback, then benchmark FireRedVAD on Netflix-GT before positioning). |
132132
| **#309** | Persistent White Screen in GUI (mklink + WebView2) | Jerry199022 | `AWAITING REPLY` | **NEW 04-26.** Detailed bug report with `C:\Users\Jerry` mklink-junctioned to D: drive. Even pure pywebview test fails with white screen → upstream pywebview/WebView2 issue, not WhisperJAV-specific. **Responded 04-30** ([comment-4358678848](https://github.com/meizhong986/WhisperJAV/issues/309#issuecomment-4358678848)): acknowledged upstream boundary, suggested workarounds (non-junctioned Windows account, VM, `WEBVIEW2_BROWSER_EXECUTABLE_FOLDER`), pointed to CLI fallback. Cluster with #225, #240. |
133133
| **#307** | Question: Principles behind JA SRT sanitization | yedkung69-ctrl | `AWAITING REPLY` | **NEW 04-23.** Polite thanks + deep question on sanitization pipeline internals. **Responded 04-30** ([comment-4358679709](https://github.com/meizhong986/WhisperJAV/issues/307#issuecomment-4358679709)): explained 3-layer pipeline (1. hallucination_remover with exact-list + regex gists, 2. generic post-processing — symbol-only/CPS/repetition, 3. model-specific cleaners e.g. AnimeWhisperCleaner), clarified VAD acoustic features are consumed by Whisper itself (not by sanitizer), pointed to source in `whisperjav/modules/subtitle_pipeline/cleaners/`. Could become a FAQ entry. |
134134
| **#308** | Add English source + ASS/SSA subtitle format | SangenBR | `NEEDS RESPONSE` | **NEW 04-23.** Feature request with screenshot: (1) add "English" to Source Language in AI SRT Translate (currently the target is English but source isn't selectable as English), (2) support .ass / .ssa input in the translate pipeline. P2 feature — tracked under Feature Requests. |

0 commit comments

Comments
 (0)