Try fixing bad live transcription performance.

src/App.tsx

@@ -297,6 +297,9 @@ export function App() {
     // live. `.en` models take neither (same rule as the file flow).
     const options = (getModel(modelId)?.multilingual ?? false) ? { language, task } : {};
     const transcriber = new LiveTranscriber({
+      tentativeIntervalMs: 2500,
+      minTentativeSeconds: 2.0,
+      segmenter: { commitSilenceSeconds: 2.5, maxSegmentSeconds: 29 },
       transcribe: (audio) => transcribeLiveSegment(worker, audio, options),
       onUpdate: (update) => {
         setLiveCommitted(update.committedText);

src/lib/mic.ts

@@ -131,7 +131,7 @@ export async function startMicCapture(
     audio: { echoCancellation: true, noiseSuppression: true, autoGainControl: true },
   });
 
-  const context = new AudioContext();
+  const context = new AudioContext({ sampleRate: TARGET_SAMPLE_RATE });
   // Contexts created outside a user gesture start suspended; the start button
   // click satisfies the gesture requirement, so this resolves immediately.
   await context.resume().catch(() => {});

src/worker/inference.worker.ts

@@ -209,7 +209,13 @@ async function transcribe(message: TranscribeMessage): Promise<void> {
   }
   const asr = await active.promise;
 
-  const streamer = new WhisperTextStreamer(
+  const options: Record<string, unknown> = {
+    chunk_length_s: 30,
+    stride_length_s: 5,
+    return_timestamps: true,
+  };
+  if (message.requestId === undefined) {
+    options.streamer = new WhisperTextStreamer(
       // The ASR pipeline for Whisper models always carries a WhisperTokenizer.
       asr.tokenizer as ConstructorParameters<typeof WhisperTextStreamer>[0],
       {
@@ -218,13 +224,7 @@ async function transcribe(message: TranscribeMessage): Promise<void> {
           post({ type: "partial", text, requestId: message.requestId }),
       },
     );
-
-  const options: Record<string, unknown> = {
-    chunk_length_s: 30,
-    stride_length_s: 5,
-    return_timestamps: true,
-    streamer,
-  };
+  }
   // Only meaningful for multilingual models; the UI omits these for `.en`
   // variants. "auto" / undefined lets Whisper detect the language.
   if (message.language !== undefined && message.language !== "auto") {