Implement live chunked STT runtime

Implement live chunked local STT runtime with chunk metadata, retry/resume support, bounded live audio queues, and progress reporting.

Author: qyinm

apps/desktop/main.cjs

@@ -2,6 +2,97 @@ const assert = require("node:assert/strict");
 
 process.env.MIRROR_NOTE_TEST_EXPORTS = "1";
 
-const { nativeHelperWorkingDirectory } = require("../main.cjs");
+const {
+  latestFailedSTTChunks,
+  latestRetryableSTTChunks,
+  nativeHelperWorkingDirectory,
+  normalizeTranscriptSegments,
+  parseSTTChunkLedgerJSONL,
+  parseTranscriptJSONL,
+  serializeTranscriptJSONL,
+} = require("../main.cjs");
 
 assert.equal(nativeHelperWorkingDirectory(), process.cwd());
+
+const runtimeSegment = {
+  id: "seg-1",
+  sessionID: "session-a",
+  chunkID: "session-a:chunk:0",
+  eventType: "final_segment",
+  retryState: "completed",
+  startTimeSeconds: 1,
+  endTimeSeconds: 2,
+  text: "hello",
+  source: "microphone",
+};
+assert.deepEqual(normalizeTranscriptSegments([runtimeSegment])[0], runtimeSegment);
+assert.deepEqual(parseTranscriptJSONL(serializeTranscriptJSONL([runtimeSegment]))[0], runtimeSegment);
+
+const ledgerRecords = parseSTTChunkLedgerJSONL([
+  JSON.stringify({
+    sessionID: "session-a",
+    chunkID: "session-a:microphone:0",
+    source: "microphone",
+    chunkIndex: 0,
+    status: "queued",
+    startTimeSeconds: 1,
+    endTimeSeconds: 2,
+    sampleCount: 16000,
+  }),
+  JSON.stringify({
+    sessionID: "session-a",
+    chunkID: "session-a:microphone:0",
+    source: "microphone",
+    chunkIndex: 0,
+    status: "failed",
+    startTimeSeconds: 1,
+    endTimeSeconds: 2,
+    sampleCount: 16000,
+    error: "boom",
+  }),
+  JSON.stringify({
+    sessionID: "session-a",
+    chunkID: "session-a:system:1",
+    source: "system",
+    chunkIndex: 1,
+    status: "completed",
+    startTimeSeconds: 3,
+    endTimeSeconds: 4,
+    sampleCount: 16000,
+    segmentCount: 1,
+  }),
+].join("\n"));
+assert.deepEqual(latestFailedSTTChunks(ledgerRecords).map((record) => record.chunkID), [
+  "session-a:microphone:0",
+]);
+assert.deepEqual(
+  latestRetryableSTTChunks(ledgerRecords, new Set(["session-a:system:1"])).map((record) => record.chunkID),
+  ["session-a:system:1"],
+);
+
+const interruptedLedgerRecords = parseSTTChunkLedgerJSONL([
+  JSON.stringify({
+    sessionID: "session-a",
+    chunkID: "session-a:microphone:2",
+    source: "microphone",
+    chunkIndex: 2,
+    status: "running",
+    startTimeSeconds: 5,
+    endTimeSeconds: 10,
+    sampleCount: 80000,
+  }),
+  JSON.stringify({
+    sessionID: "session-a",
+    chunkID: "session-a:system:3",
+    source: "system",
+    chunkIndex: 3,
+    status: "queued",
+    startTimeSeconds: 12,
+    endTimeSeconds: 17,
+    sampleCount: 80000,
+  }),
+].join("\n"));
+assert.deepEqual(latestRetryableSTTChunks(interruptedLedgerRecords, new Set()).map((record) => record.chunkID), [
+  "session-a:microphone:2",
+  "session-a:system:3",
+]);

apps/desktop/main/native-launch.node-test.cjs

@@ -704,8 +704,18 @@ export function App() {
       const percentLabel = Number.isFinite(percentage)
         ? `${Math.round(Math.max(0, Math.min(1, percentage)) * 100)}%`
         : "";
+      const completedChunks = Number(event.transcriptionProgress.completedChunkCount || 0);
+      const runningChunks = Number(event.transcriptionProgress.runningChunkCount || 0);
+      const failedChunks = Number(event.transcriptionProgress.failedChunkCount || 0);
+      const chunkLabel = completedChunks > 0 || runningChunks > 0 || failedChunks > 0
+        ? ` · ${t("status.chunkProgress", {
+          completed: completedChunks,
+          running: runningChunks,
+          failed: failedChunks,
+        })}`
+        : "";
       const finalizingTranscript = t("status.finalizingTranscript");
-      setCaptureDetail(percentLabel ? `${finalizingTranscript} ${percentLabel}` : finalizingTranscript);
+      setCaptureDetail(percentLabel ? `${finalizingTranscript} ${percentLabel}${chunkLabel}` : `${finalizingTranscript}${chunkLabel}`);
       return;
     }
     if (event.kind === "finished") {

apps/desktop/src/App.tsx

@@ -79,6 +79,7 @@ type I18nKey =
   | "sidebar.templates"
   | "status.creatingNote"
   | "status.deletingNote"
+  | "status.chunkProgress"
   | "status.finalizingTranscript"
   | "status.generatingSummary"
   | "status.loadingNotes"
@@ -246,6 +247,7 @@ const translations: Record<AppLanguage, Record<I18nKey, string>> = {
     "sidebar.templates": "Templates",
     "status.creatingNote": "Creating note...",
     "status.deletingNote": "Deleting note...",
+    "status.chunkProgress": "chunks {completed} done, {running} running, {failed} failed",
     "status.finalizingTranscript": "Finalizing transcript...",
     "status.generatingSummary": "Generating summary...",
     "status.loadingNotes": "Loading notes...",
@@ -412,6 +414,7 @@ const translations: Record<AppLanguage, Record<I18nKey, string>> = {
     "sidebar.templates": "템플릿",
     "status.creatingNote": "노트 생성 중...",
     "status.deletingNote": "노트 삭제 중...",
+    "status.chunkProgress": "청크 {completed} 완료, {running} 실행 중, {failed} 실패",
     "status.finalizingTranscript": "전사 마무리 중...",
     "status.generatingSummary": "요약 생성 중...",
     "status.loadingNotes": "노트 불러오는 중...",

apps/desktop/src/i18n.tsx

@@ -180,6 +180,10 @@ export type CaptureEvent =
         processedSeconds?: number;
         totalSeconds?: number;
         percentage?: number;
+        queuedChunkCount?: number;
+        runningChunkCount?: number;
+        completedChunkCount?: number;
+        failedChunkCount?: number;
       };
     }
   | { kind: "finished" };