Transcription: don't pass task/language to English-only models

Whisper ".en" models reject `task`/`language`; transformers.js throws
"Cannot specify `task` or `language` for an English-only model" if either
is passed — and the web engine passed `task: 'transcribe'` (plus a
possibly-undefined `language`) unconditionally. Since the default model is
tiny.en, web/PWA transcription failed for every default user.

Only add `task`/`language` for multilingual models now (translation also
requires a multilingual model), and never pass an explicit `undefined`
language (which trips the same check). Mirror the same gate on native
(whisper.rn) so a .en model is never asked to do a language/translate it
can't.

tsc clean, 279 tests pass, web export OK.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

src/lib/transcription/engineImpl.native.ts

@@ -196,11 +196,15 @@ export const engine: TranscriptionEngine = {
     // Float32Array whose `.buffer` is exactly these samples.
     const pcm = audio.samples.slice();
 
+    // English-only models (".en") only do English — only pass language/translate
+    // for multilingual models (mirrors the web engine; keeps whisper.cpp from
+    // being asked to do something a .en model can't).
+    const multilingual = MODELS[opts.modelId].multilingual;
     // transcribeData returns { stop, promise }; await the promise.
     const { promise } = ctx.transcribeData(pcm.buffer, {
       // language undefined => whisper.rn auto-detects ('auto').
-      language: opts.language,
-      translate: opts.translate ?? false,
+      ...(multilingual && opts.language ? { language: opts.language } : {}),
+      ...(multilingual ? { translate: opts.translate ?? false } : {}),
     });
     const result = await promise;
 

src/lib/transcription/engineImpl.web.ts

@@ -117,13 +117,22 @@ export const engine: TranscriptionEngine = {
     const asr = loaded.get(opts.modelId);
     if (!asr) throw new Error(`Model "${opts.modelId}" is not loaded; call loadModel() first.`);
 
-    const out = await asr(audio.samples, {
+    // English-only Whisper models (".en") reject `language`/`task` — transformers.js
+    // throws "Cannot specify `task` or `language` for an English-only model" if
+    // either is passed. Only multilingual models accept them (translation also
+    // requires a multilingual model). So we add those keys ONLY when multilingual,
+    // and never pass an explicit `undefined` language (which trips the same check).
+    const genOpts: Record<string, unknown> = {
       return_timestamps: true,
       // One window at a time; 30s matches Whisper's frame so it won't re-chunk.
       chunk_length_s: 30,
-      language: opts.language,
-      task: opts.translate ? 'translate' : 'transcribe',
-    });
+    };
+    if (MODELS[opts.modelId].multilingual) {
+      genOpts.task = opts.translate ? 'translate' : 'transcribe';
+      if (opts.language) genOpts.language = opts.language;
+    }
+
+    const out = await asr(audio.samples, genOpts);
 
     const segments: Segment[] = [];
     for (const c of out.chunks ?? []) {