Tune realtime STT defaults

src/index.ts

@@ -69,8 +69,9 @@ async function runLoopback(): Promise<void> {
   await stt.warmup();
   logger.info("STT warmup finished");
 
-  const transcriptionQueue: Buffer[] = [];
+  const transcriptionQueue: Array<{ pcm16: Buffer; queuedAt: number; index: number }> = [];
   let transcribing = false;
+  let nextSegmentIndex = 1;
 
   const runNext = async (): Promise<void> => {
     if (transcribing) {
@@ -83,11 +84,17 @@ async function runLoopback(): Promise<void> {
 
     transcribing = true;
     try {
-      const text = await stt.transcribePcm16(next);
+      const startedAt = Date.now();
+      const text = await stt.transcribePcm16(next.pcm16);
+      logger.info("STT latency", {
+        index: next.index,
+        wait_ms: startedAt - next.queuedAt,
+        transcribe_ms: Date.now() - startedAt,
+      });
       if (!text) {
         logger.info("빈 전사 결과");
       } else {
-        logger.info("Transcript", text);
+        logger.info("Transcript", { index: next.index, text });
         if (config.DEBUG_TRANSCRIPTS) {
           console.log(`\n[text] ${text}\n`);
         }
@@ -101,6 +108,13 @@ async function runLoopback(): Promise<void> {
   };
 
   const segmenter = new RealtimeSegmenter({
+    preRollSamples: config.SEGMENT_PREROLL_SAMPLES,
+    speechStartThreshold: config.SEGMENT_START_THRESHOLD,
+    speechContinueThreshold: config.SEGMENT_CONTINUE_THRESHOLD,
+    speechStartFrames: config.SEGMENT_START_FRAMES,
+    speechEndFrames: config.SEGMENT_END_FRAMES,
+    minSpeechSamples: config.SEGMENT_MIN_SPEECH_SAMPLES,
+    maxSpeechSamples: config.SEGMENT_MAX_SPEECH_SAMPLES,
     onLevel: (peak) => {
       if (peak > maxPeak) {
         maxPeak = peak;
@@ -128,11 +142,17 @@ async function runLoopback(): Promise<void> {
     },
     onSpeechReady: (samples) => {
       emittedSegmentCount += 1;
-      logger.info("Speech segment ready", { index: emittedSegmentCount, samples });
+      logger.info("Speech segment ready", { index: emittedSegmentCount, samples, ms: Math.round((samples / 16000) * 1000) });
     },
     onSegment: (pcm16) => {
-      transcriptionQueue.push(pcm16);
+      const index = nextSegmentIndex++;
+      transcriptionQueue.push({
+        pcm16,
+        queuedAt: Date.now(),
+        index,
+      });
       logger.info("Queued segment for STT", {
+        index,
         queue: transcriptionQueue.length,
         bytes: pcm16.length,
       });
@@ -167,6 +187,7 @@ async function runLoopback(): Promise<void> {
   console.log(`source: ${config.AUDIO_SOURCE ?? "unset"}`);
   console.log(`model: ${config.WHISPER_MODEL}`);
   console.log(`language: ${config.WHISPER_LANGUAGE}`);
+  console.log(`beam: ${config.WHISPER_BEAM_SIZE}`);
 
   setInterval(() => {
     const now = Date.now();