@absolutejs/voice 0.0.22-beta.641 → 0.0.22-beta.642

package/dist/index.js

@@ -4297,6 +4297,7 @@ var createVoiceSession = (options) => {
   let lastSttReconnectAt = 0;
   let lastSpeechEnergyAt = 0;
   let sttHealthPhaseStart = 0;
+  let sttDeafConfirmedAt = 0;
   let lastSttRecoverySpokenAt = 0;
   let activeTTSTurnId;
   let assistantSpeechEndsAt = 0;
@@ -4431,30 +4432,21 @@ var createVoiceSession = (options) => {
       stuckCloseFired = true;
       const reason = `no caller progress for ${stuckCloseAfterMs}ms`;
       const { lastSpeechAt, lastTranscriptAt } = snapshot.currentTurn;
-      const speechWithoutTranscript = lastSpeechAt !== undefined && (lastTranscriptAt === undefined || lastSpeechAt > lastTranscriptAt);
+      const sttDeafConfirmed = sttDeafConfirmedAt > 0 && (lastTranscriptAt === undefined || sttDeafConfirmedAt > lastTranscriptAt);
       const diagnostic = {
         action: "stuck-call-close",
-        classification: speechWithoutTranscript ? "stt-deaf" : "caller-idle",
+        classification: sttDeafConfirmed ? "stt-deaf" : "caller-idle",
         lastAudioAt: snapshot.currentTurn.lastAudioAt,
         lastSpeechAt,
         lastTranscriptAt,
         reason,
         turnCount: snapshot.turns.length
       };
-      if (speechWithoutTranscript) {
-        const error = `caller speech received but STT produced no transcript for ${stuckCloseAfterMs}ms`;
-        await appendTrace({
-          payload: { ...diagnostic, error },
-          session: snapshot,
-          type: "session.error"
-        });
-      } else {
-        await appendTrace({
-          payload: diagnostic,
-          session: snapshot,
-          type: "call.lifecycle"
-        });
-      }
+      await appendTrace({
+        payload: diagnostic,
+        session: snapshot,
+        type: "call.lifecycle"
+      });
       if (stuckCloseConfig?.line) {
         await speakResolvedLine(stuckCloseConfig.line, snapshot);
       }
@@ -5675,6 +5667,7 @@ var createVoiceSession = (options) => {
     };
   };
   const handlePartial = async (transcript) => {
+    sttDeafConfirmedAt = 0;
     if (transcript.text.trim()) {
       kickStuckCloseWatchdog();
       kickIdleRepromptWatchdog(true);
@@ -5773,6 +5766,7 @@ var createVoiceSession = (options) => {
   };
   const handleFinal = async (transcript) => {
     sttReconnectCount = 0;
+    sttDeafConfirmedAt = 0;
     if (backchannelBargeInGuard && backchannelSuppressedAt !== null && Date.now() - backchannelSuppressedAt < BACKCHANNEL_DROP_WINDOW_MS && isBackchannelUtterance(transcript.text.trim())) {
       backchannelSuppressedAt = null;
       return;
@@ -6987,6 +6981,7 @@ var createVoiceSession = (options) => {
       lastSpeechEnergyAt = nowMs;
       const lastTranscriptAt = latest.currentTurn.lastTranscriptAt ?? 0;
       if (!options.realtime && sttSession && lastTranscriptAt < sttHealthPhaseStart && nowMs - sttHealthPhaseStart >= STT_HEALTH_STALE_MS) {
+        sttDeafConfirmedAt = nowMs;
         sttReconnectCount = nowMs - lastSttReconnectAt < STT_RECONNECT_FLAP_WINDOW_MS ? sttReconnectCount + 1 : 1;
         lastSttReconnectAt = nowMs;
         sttHealthPhaseStart = nowMs;
@@ -6996,6 +6991,7 @@ var createVoiceSession = (options) => {
             payload: {
               action: "stt-health-reconnect",
               attempt: sttReconnectCount,
+              classification: "stt-deaf",
               reason: `no transcript for ${STT_HEALTH_STALE_MS}ms of continuous speech`
             },
             session: latest,

package/dist/testing/index.js

@@ -328,7 +328,12 @@ var applyVoiceAudioCondition = (fixture, condition) => {
     for (let index = 0;index < output.length; index += 1)
       output[index] = clamp(output[index] + random() * Math.sqrt(3) * noiseRms);
   }
-  return { ...fixture, audio: bytes(output), id: `${fixture.id}--${condition.id}`, tags: [...fixture.tags ?? [], "conditioned", condition.id] };
+  return {
+    ...fixture,
+    audio: bytes(output),
+    id: `${fixture.id}--${condition.id}`,
+    tags: [...fixture.tags ?? [], "conditioned", condition.id]
+  };
 };
 var buildVoiceAudioMatrix = (fixtures, conditions) => fixtures.flatMap((fixture) => conditions.map((condition) => applyVoiceAudioCondition(fixture, condition)));
 // src/testing/stt.ts
@@ -6005,9 +6010,15 @@ var stableBenchmarkJson = (value) => {
 };
 var buildVoiceBenchmarkArtifact = (manifest, report) => {
   const payload = { manifest, report };
-  return { ...payload, artifactSha256: sha256Bytes(stableBenchmarkJson(payload)) };
+  return {
+    ...payload,
+    artifactSha256: sha256Bytes(stableBenchmarkJson(payload))
+  };
 };
-var verifyVoiceBenchmarkArtifact = (artifact) => artifact.artifactSha256 === sha256Bytes(stableBenchmarkJson({ manifest: artifact.manifest, report: artifact.report }));
+var verifyVoiceBenchmarkArtifact = (artifact) => artifact.artifactSha256 === sha256Bytes(stableBenchmarkJson({
+  manifest: artifact.manifest,
+  report: artifact.report
+}));
 // src/core/memoryStore.ts
 var createVoiceMemoryStore = () => {
   const sessions = new Map;
@@ -6929,6 +6940,7 @@ var createVoiceSession = (options) => {
   let lastSttReconnectAt = 0;
   let lastSpeechEnergyAt = 0;
   let sttHealthPhaseStart = 0;
+  let sttDeafConfirmedAt = 0;
   let lastSttRecoverySpokenAt = 0;
   let activeTTSTurnId;
   let assistantSpeechEndsAt = 0;
@@ -7063,30 +7075,21 @@ var createVoiceSession = (options) => {
       stuckCloseFired = true;
       const reason = `no caller progress for ${stuckCloseAfterMs}ms`;
       const { lastSpeechAt, lastTranscriptAt } = snapshot.currentTurn;
-      const speechWithoutTranscript = lastSpeechAt !== undefined && (lastTranscriptAt === undefined || lastSpeechAt > lastTranscriptAt);
+      const sttDeafConfirmed = sttDeafConfirmedAt > 0 && (lastTranscriptAt === undefined || sttDeafConfirmedAt > lastTranscriptAt);
       const diagnostic = {
         action: "stuck-call-close",
-        classification: speechWithoutTranscript ? "stt-deaf" : "caller-idle",
+        classification: sttDeafConfirmed ? "stt-deaf" : "caller-idle",
         lastAudioAt: snapshot.currentTurn.lastAudioAt,
         lastSpeechAt,
         lastTranscriptAt,
         reason,
         turnCount: snapshot.turns.length
       };
-      if (speechWithoutTranscript) {
-        const error = `caller speech received but STT produced no transcript for ${stuckCloseAfterMs}ms`;
-        await appendTrace({
-          payload: { ...diagnostic, error },
-          session: snapshot,
-          type: "session.error"
-        });
-      } else {
-        await appendTrace({
-          payload: diagnostic,
-          session: snapshot,
-          type: "call.lifecycle"
-        });
-      }
+      await appendTrace({
+        payload: diagnostic,
+        session: snapshot,
+        type: "call.lifecycle"
+      });
       if (stuckCloseConfig?.line) {
         await speakResolvedLine(stuckCloseConfig.line, snapshot);
       }
@@ -8307,6 +8310,7 @@ var createVoiceSession = (options) => {
     };
   };
   const handlePartial = async (transcript) => {
+    sttDeafConfirmedAt = 0;
     if (transcript.text.trim()) {
       kickStuckCloseWatchdog();
       kickIdleRepromptWatchdog(true);
@@ -8405,6 +8409,7 @@ var createVoiceSession = (options) => {
   };
   const handleFinal = async (transcript) => {
     sttReconnectCount = 0;
+    sttDeafConfirmedAt = 0;
     if (backchannelBargeInGuard && backchannelSuppressedAt !== null && Date.now() - backchannelSuppressedAt < BACKCHANNEL_DROP_WINDOW_MS && isBackchannelUtterance(transcript.text.trim())) {
       backchannelSuppressedAt = null;
       return;
@@ -9619,6 +9624,7 @@ var createVoiceSession = (options) => {
       lastSpeechEnergyAt = nowMs;
       const lastTranscriptAt = latest.currentTurn.lastTranscriptAt ?? 0;
       if (!options.realtime && sttSession && lastTranscriptAt < sttHealthPhaseStart && nowMs - sttHealthPhaseStart >= STT_HEALTH_STALE_MS) {
+        sttDeafConfirmedAt = nowMs;
         sttReconnectCount = nowMs - lastSttReconnectAt < STT_RECONNECT_FLAP_WINDOW_MS ? sttReconnectCount + 1 : 1;
         lastSttReconnectAt = nowMs;
         sttHealthPhaseStart = nowMs;
@@ -9628,6 +9634,7 @@ var createVoiceSession = (options) => {
             payload: {
               action: "stt-health-reconnect",
               attempt: sttReconnectCount,
+              classification: "stt-deaf",
               reason: `no transcript for ${STT_HEALTH_STALE_MS}ms of continuous speech`
             },
             session: latest,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@absolutejs/voice",
-  "version": "0.0.22-beta.641",
+  "version": "0.0.22-beta.642",
   "description": "Voice primitives and Elysia plugin for AbsoluteJS",
   "repository": {
     "type": "git",