Merge feat/longform-pipeline: VAD long-form pipeline, opt-in (spec 001 workstream 3)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

# Conflicts:
#	docs/Architecture.md
#	docs/Decisions.md
#	src/App.tsx
#	src/index.css
#	src/lib/engine.ts
#	src/worker/inference.worker.ts
#	src/worker/protocol.ts
#	tests/engine.test.ts
#	tests/live-real-model.test.ts
#	tests/offline.md
#	tests/protocol.test.ts

docs/Architecture.md

@@ -84,6 +84,9 @@ src/
     translate.worker.ts    # transcript translation (Opus-MT), its own worker + ort instance (D-WS4b)
     translateProtocol.ts   # typed messages for the translate worker
     opusMt.ts              # Opus-MT loader shared by the worker, bench script and real-model test
+    longform/              # long-form file pipeline (D-WS3a): pure vad.ts / quality.ts / pipeline.ts
+                           # behind transcribeLongForm(audio, deps, options); transformers.ts implements
+                           # the deps (Silero VAD + Whisper decodes) on the same Transformers.js runtime
   pwa/
     sw.ts                  # service worker (app-shell precache, /ort runtime cache)
     precache.ts            # placeholder-injection helpers shared by build + dev serve
@@ -161,6 +164,22 @@ environment). See `WorkBreakdown.md` for the itemized checklist.)
    to the decoder input of every 30 s window by a wrapper around the model's
    `generate()`, which also strips it from the output again (D-WS2a).
    Live passes (4.1b) carry the same `prompt`.
+6. **Long-form branch (D-WS3a/D-WS3b, "Advanced → Skip silence", off by default).**
+   With the option on, every load posted for a file run or the Download button carries
+   `vad: true`; once the Whisper pipeline is ready the worker loads Silero VAD
+   (`onnx-community/silero-vad`, 2.2 MB, Cache API like the weights; offline it is
+   used only if already cached). A file pass with `longForm: true` then runs
+   `transcribeLongForm` (`worker/longform/`): VAD speech probabilities over the
+   whole buffer → speech spans (faster-whisper `vad.py` port) → chunks ≤ 30 s cut on
+   speech boundaries → one Whisper decode per chunk (language detected once on the
+   first chunk and pinned for multilingual models) → timestamps mapped back to the
+   file's timeline; progress follows the variable chunks
+   (`TranscribeProgressTracker.forWindows`). Same `partial` /
+   `transcribe-progress` / `result` messages as step 5. `choosePipeline` keeps the
+   chunked path whenever the VAD model isn't loaded, and live passes never take
+   this branch (live keeps its RMS segmenter, D-016/D-024). A finished transcript
+   with no words shows "No speech was found in this recording." The names-and-terms
+   hint (`prompt`) is applied to every chunk's decode the same way (D-WS2a).
 
 ### 4.1b Live microphone transcription (anonymous, browser engine — D-016)
 1. `LiveControls` starts a session: `getUserMedia` → `AudioContext` at the
@@ -224,7 +243,9 @@ environment). See `WorkBreakdown.md` for the itemized checklist.)
   large to precache eagerly.
 - Model weights live in the Cache API (`transformers-cache`) via Transformers.js —
   independent of the SW, which never touches that cache (D-005). `/api/*` is
-  network-only.
+  network-only. The Silero VAD model of the long-form branch lives there too
+  (D-WS3b); offline, the worker checks that cache before loading it and never
+  fetches, falling back to the chunked path if it is missing.
 - Updates: `pwa/register.ts` surfaces a waiting SW as a toast; accepting posts
   `SKIP_WAITING`, activation drops old `whisper-*` caches, and the page reloads.
 - Result: with the app installed and at least one model previously loaded, the full

docs/Decisions.md

@@ -1619,3 +1619,87 @@ itself and accepts only text, so the sidecar passes it the budget-cut ids
 decoded back to text (re-encoding gives the same ids, checked with the real
 tokenizer incl. "Århus, Ærø"); server-live passes keep the ids. Pinned by the
 stubbed-sidecar test "the vocabulary hint reaches the batched pipeline as text".
+
+## D-WS3a — Long-form file pipeline (Silero VAD chunks, language pinned once) ships as an Advanced option, off by default
+**Status:** Accepted (2026-10-30) — spec 001 Goal 3
+Browser file transcription cuts audio into fixed 30 s windows with 5 s
+overlap (Transformers.js's chunked pipeline), decoding silence and
+overlaps, and with "Auto-detect" Transformers.js 4.2 silently decodes as
+English.
+**Decision:** `src/worker/longform/` adds `transcribeLongForm(audio, deps,
+options)`: Silero VAD speech probabilities over the whole buffer → speech
+spans (line-for-line port of faster-whisper 1.2.1 `get_speech_timestamps`,
+parity-checked against the Python on 20 000 random windows) → chunks
+≤ 30 s cut on speech boundaries (`collect_chunks`) → one Whisper decode per
+chunk → timestamps mapped back to the file's timeline. For a multilingual
+model without a language, it detects once on the first speech chunk (one
+decoder step on that chunk's encoder output, which the decode then reuses)
+and pins it for every chunk. Defaults: VAD `threshold 0.5`,
+`min_silence 2000 ms`, `speech_pad 400 ms`, `max_speech 30 s`; repetition
+guard **off** (`no_repeat_ngram_size 0`, `repetition_penalty 1`);
+temperature fallback (compression ratio > 2.4 / avg log-prob < −1.0 →
+0.2…1.0) implemented but **off**. Inference is injected (`LongFormDeps`),
+so the 90-minute case is unit-tested: VAD gets the caller's buffer, decodes
+get ≤ 30 s copies. Audio with no speech returns an empty transcript without
+a decode; the UI shows "No speech was found in this recording."
+**Measurements** (onnxruntime-node CPU, Large v3 Turbo q4 = the browser's
+WASM weights; machine shared with six other workstreams, old/new run
+back-to-back per clip; WER vs `tests/audio_files/*.txt`):
+
+| clip | old s | long-form s | old WER | long-form WER | old / lf tokens |
+|---|---|---|---|---|---|
+| en_aiproblem (14 s) | 3.6 | 3.3 | 0.000 | 0.000 | 47 / 47 |
+| da_festuge, `da` (21 s) | 3.4 · 6.3 | 3.5 · 6.2 | 0.141 | 0.141 | 122 / 121 |
+| da_festuge, auto | 6.7 | 4.7 | 0.938 (English) | 0.141 (`da` detected) | 88 / 121 |
+| en_alien (15 min) | 210.2 | 115.0 | 0.255 | 0.140 | 3933 / 2258 |
+| meeting (**synthetic**, 2 min 45 s) | 41.9 | 37.5 | 0.0317 | 0.0422 | 776 / 594 |
+
+(da_festuge with `da` was measured twice, in two load conditions. The
+meeting is `tests/meetingFixture.ts`: en_aiproblem and three en_alien
+excerpts joined by 3–8 s pauses of −60 dBFS noise — **synthetic**. Its
+long-form errors are mostly spelling of casual speech: "gonna" for
+"going to", "killin", "livestreamed".)
+**Rollout (spec rule: default only if ≤ old WER + 0.01 and faster on every
+clip):** not met. The synthetic meeting is +0.0106 WER (just over the
+tolerance) and da_festuge with an explicit language is a tie within
+noise (3.5 vs 3.4 s, then 6.2 vs 6.3 s). So the old chunked path stays
+the default and long-form ships as **Advanced → "Skip silence in uploaded
+files"**, off by default, remembered in localStorage (`whisper-long-form`).
+It is the only path whose "Auto-detect" actually detects, and roughly
+halves time and WER on the long, pause-heavy en_alien.
+The repetition guard and fallback variants were not measured (heavy runs
+kept to one baseline + one long-form run per clip); neither is on, so no
+unmeasured setting ships as a default.
+**Consequence:** live mode is untouched (its RMS segmenter, D-016/D-024).
+`tests/live-real-model.test.ts` now asserts whole-file WER of the chunked
+path and of long-form (explicit and auto language) against the per-clip
+bars. Whisper's `generate()` drops kwargs not in `forward_params`, so
+`createWhisperDeps` adds `encoder_outputs` to that list on its pipeline
+instance; a Transformers.js upgrade must keep the "one encoder pass per
+chunk" assertion green. WebGPU timings are unmeasured.
+
+**Integration note (2026-10-30, merge with D-WS2a):** the long-form adapter
+(`worker/longform/transformers.ts`) applies the names-and-terms hint to every
+chunk's decode with the same decoder-prompt prefix as the chunked path
+(tokenized once per pass); before the merge the `prompt` option reached the
+adapter but was ignored. Pinned by a unit test in
+`tests/longform-transformers.test.ts`.
+
+## D-WS3b — Silero VAD model: onnx-community/silero-vad from the Hub, cached like Whisper weights, never fetched offline
+**Status:** Accepted (2026-10-30) — spec 001 Goal 3
+**Decision:** the long-form pipeline's VAD is Silero VAD v5 as exported by
+`onnx-community/silero-vad` (`onnx/model.onnx`, fp32, 2 243 022 bytes, MIT;
+commit `e71cae9` on 2026-10-30). It loads through Transformers.js's
+`AutoModel` as a bare ONNX graph (`model_type: "custom"`) on the same
+onnxruntime as Whisper (WASM device — a 2 MB model gains nothing from a
+GPU), fetched from the Hub like the weights (CORS, so COEP allows it) and
+stored in the same Cache API store (D-005). Windows of 512 samples with 64
+samples of context and the `[2, 1, 128]` state, as in Silero's own
+wrapper. It is loaded only with "Skip silence" on, after the Whisper
+pipeline is ready, on the same user-initiated load (D-020). Offline
+(`navigator.onLine === false`) it loads only if already cached and never
+fetches; missing or failed, the worker logs it and files take the chunked
+path (`choosePipeline`). Live mode keeps its energy gate (D-016).
+**Rejected:** bundling the model in `dist/` (2 MB on every install for an
+off-by-default option); running it on the main thread (D-013).
+**Consequence:** the first "Skip silence" run online downloads 2.2 MB.

docs/WorkBreakdown.md

@@ -423,6 +423,16 @@ Found in spec 001 workstream 6 (D-WS6a), open:
       (D-WS6a). Judging them needs a longer, harder Danish reference clip than
       this repo has.
 
+Found while building the long-form pipeline (spec 001 Goal 3, D-WS3a), 2026-10-29:
+- [ ] "Auto-detect" on the default (chunked) file path does not detect: with no
+      `language`, Transformers.js 4.2 logs "No language specified - defaulting to
+      English" and decodes Danish audio as an English translation (da_festuge WER
+      0.94 vs 0.14 with `da`). Long-form ("Skip silence") detects once and pins
+      the language; the chunked path is unchanged.
+- [ ] Spec 001's measurement fallback mentions a hint list of the names in
+      `en_alien`'s reference; `tests/meetingFixture.ts` returns audio + reference
+      only. Add the list when the vocabulary hint (workstream 2) lands.
+
 ## Recurring
 - [ ] Maintain `AGENTS.md`:
   - **Add a rule only in response to a real failure.** When an agent repeatedly