@absolutejs/voice 0.0.22-beta.580 → 0.0.22-beta.581

package/dist/index.js

@@ -3936,6 +3936,8 @@ var createVoiceSession = (options) => {
   let activeAdapterGeneration = 0;
   let activeTTSTurnId;
   let assistantSpeechEndsAt = 0;
+  let lastAssistantAudioAt = 0;
+  let lastTtsSendAt = 0;
   let fillerTimer = null;
   let fillerActive = false;
   let fillerToken = 0;
@@ -4209,6 +4211,15 @@ var createVoiceSession = (options) => {
     });
     return result;
   };
+  let assistantAudioQueue = Promise.resolve();
+  const runAudioSerial = (operation) => {
+    const next = assistantAudioQueue.then(operation);
+    assistantAudioQueue = next.then(() => {
+      return;
+    }, () => {
+      return;
+    });
+  };
   const closeAdapter = async (reason) => {
     if (!sttSession) {
       return;
@@ -4377,6 +4388,7 @@ var createVoiceSession = (options) => {
       const chunkMs = normalizedChunk.byteLength / bytesPerSecond * 1000;
       assistantSpeechEndsAt = Math.max(assistantSpeechEndsAt, Date.now()) + chunkMs;
     }
+    lastAssistantAudioAt = Date.now();
     if (activeTTSTurnId) {
       await appendTurnLatencyStage({
         at: input.receivedAt,
@@ -4486,18 +4498,28 @@ var createVoiceSession = (options) => {
       session
     });
   };
-  const DRAIN_POLL_MS = 200;
+  const DRAIN_POLL_MS = 100;
   const DRAIN_TAIL_BUFFER_MS = 300;
-  const DRAIN_MAX_MS = 12000;
-  const drainAssistantSpeech = async () => {
+  const DRAIN_QUIET_MS = 600;
+  const DRAIN_RENDER_START_MS = 4000;
+  const DRAIN_MAX_MS = 20000;
+  const drainAssistantSpeech = async (renderPendingSince) => {
     const startedAt = Date.now();
+    const sleep3 = (delayMs) => new Promise((resolve) => {
+      setTimeout(resolve, delayMs);
+    });
     while (Date.now() - startedAt < DRAIN_MAX_MS) {
-      const remaining = assistantSpeechEndsAt + DRAIN_TAIL_BUFFER_MS - Date.now();
-      if (remaining <= 0)
+      const now = Date.now();
+      const renderStarted = lastAssistantAudioAt >= renderPendingSince || now - renderPendingSince >= DRAIN_RENDER_START_MS;
+      if (!renderStarted) {
+        await sleep3(DRAIN_POLL_MS);
+        continue;
+      }
+      const streamQuiet = now - lastAssistantAudioAt >= DRAIN_QUIET_MS;
+      const playbackDrained = assistantSpeechEndsAt + DRAIN_TAIL_BUFFER_MS <= now;
+      if (streamQuiet && playbackDrained)
         return;
-      await new Promise((resolve) => {
-        setTimeout(resolve, Math.min(remaining, DRAIN_POLL_MS));
-      });
+      await sleep3(DRAIN_POLL_MS);
     }
   };
   const completeInternal = async (result, input = {}) => {
@@ -4534,7 +4556,8 @@ var createVoiceSession = (options) => {
       return;
     }
     if (disposition === "completed") {
-      await drainAssistantSpeech();
+      await drainAssistantSpeech(lastTtsSendAt);
+      await assistantAudioQueue;
     }
     await appendTrace({
       payload: {
@@ -5204,7 +5227,10 @@ var createVoiceSession = (options) => {
     });
     if (options.realtime) {
       openedSession.on("audio", ({ chunk, format, receivedAt }) => {
-        runAdapterEvent("adapter.audio", async () => {
+        runAudioSerial(async () => {
+          if (activeAdapterGeneration !== generation) {
+            return;
+          }
           await sendAssistantAudio(chunk, {
             format,
             receivedAt
@@ -5233,7 +5259,7 @@ var createVoiceSession = (options) => {
       });
       ttsSession = openedSession;
       openedSession.on("audio", ({ chunk, format, receivedAt }) => {
-        runSerial("tts.audio", async () => {
+        runAudioSerial(async () => {
           if (ttsSession !== openedSession) {
             return;
           }
@@ -5361,6 +5387,7 @@ var createVoiceSession = (options) => {
         try {
           await ttsSession2.send(text);
           charsSent += text.length;
+          lastTtsSendAt = Date.now();
         } catch (error) {
           logger.warn("voice assistant audio send failed", {
             error: toError(error).message,
@@ -5643,6 +5670,7 @@ var createVoiceSession = (options) => {
             turnId: turn.id
           });
           await activeTTSSession.send(output.assistantText);
+          lastTtsSendAt = Date.now();
           if (options.costAccountant) {
             options.costAccountant.recordTTS({
               characters: output.assistantText.length
@@ -6067,10 +6095,12 @@ var createVoiceSession = (options) => {
         if (greetingTTSSession) {
           activeTTSTurnId = greetingTurnId;
           await greetingTTSSession.send(greetingText);
+          lastTtsSendAt = Date.now();
         } else if (options.realtime) {
           const greetingRealtimeSession = await ensureAdapter();
           activeTTSTurnId = greetingTurnId;
           await greetingRealtimeSession.send(greetingText);
+          lastTtsSendAt = Date.now();
         }
       } catch {}
     }

package/dist/testing/index.js

@@ -6053,6 +6053,8 @@ var createVoiceSession = (options) => {
   let activeAdapterGeneration = 0;
   let activeTTSTurnId;
   let assistantSpeechEndsAt = 0;
+  let lastAssistantAudioAt = 0;
+  let lastTtsSendAt = 0;
   let fillerTimer = null;
   let fillerActive = false;
   let fillerToken = 0;
@@ -6326,6 +6328,15 @@ var createVoiceSession = (options) => {
     });
     return result;
   };
+  let assistantAudioQueue = Promise.resolve();
+  const runAudioSerial = (operation) => {
+    const next = assistantAudioQueue.then(operation);
+    assistantAudioQueue = next.then(() => {
+      return;
+    }, () => {
+      return;
+    });
+  };
   const closeAdapter = async (reason) => {
     if (!sttSession) {
       return;
@@ -6494,6 +6505,7 @@ var createVoiceSession = (options) => {
       const chunkMs = normalizedChunk.byteLength / bytesPerSecond * 1000;
       assistantSpeechEndsAt = Math.max(assistantSpeechEndsAt, Date.now()) + chunkMs;
     }
+    lastAssistantAudioAt = Date.now();
     if (activeTTSTurnId) {
       await appendTurnLatencyStage({
         at: input.receivedAt,
@@ -6603,18 +6615,28 @@ var createVoiceSession = (options) => {
       session
     });
   };
-  const DRAIN_POLL_MS = 200;
+  const DRAIN_POLL_MS = 100;
   const DRAIN_TAIL_BUFFER_MS = 300;
-  const DRAIN_MAX_MS = 12000;
-  const drainAssistantSpeech = async () => {
+  const DRAIN_QUIET_MS = 600;
+  const DRAIN_RENDER_START_MS = 4000;
+  const DRAIN_MAX_MS = 20000;
+  const drainAssistantSpeech = async (renderPendingSince) => {
     const startedAt = Date.now();
+    const sleep2 = (delayMs) => new Promise((resolve2) => {
+      setTimeout(resolve2, delayMs);
+    });
     while (Date.now() - startedAt < DRAIN_MAX_MS) {
-      const remaining = assistantSpeechEndsAt + DRAIN_TAIL_BUFFER_MS - Date.now();
-      if (remaining <= 0)
+      const now = Date.now();
+      const renderStarted = lastAssistantAudioAt >= renderPendingSince || now - renderPendingSince >= DRAIN_RENDER_START_MS;
+      if (!renderStarted) {
+        await sleep2(DRAIN_POLL_MS);
+        continue;
+      }
+      const streamQuiet = now - lastAssistantAudioAt >= DRAIN_QUIET_MS;
+      const playbackDrained = assistantSpeechEndsAt + DRAIN_TAIL_BUFFER_MS <= now;
+      if (streamQuiet && playbackDrained)
         return;
-      await new Promise((resolve2) => {
-        setTimeout(resolve2, Math.min(remaining, DRAIN_POLL_MS));
-      });
+      await sleep2(DRAIN_POLL_MS);
     }
   };
   const completeInternal = async (result, input = {}) => {
@@ -6651,7 +6673,8 @@ var createVoiceSession = (options) => {
       return;
     }
     if (disposition === "completed") {
-      await drainAssistantSpeech();
+      await drainAssistantSpeech(lastTtsSendAt);
+      await assistantAudioQueue;
     }
     await appendTrace({
       payload: {
@@ -7321,7 +7344,10 @@ var createVoiceSession = (options) => {
     });
     if (options.realtime) {
       openedSession.on("audio", ({ chunk, format, receivedAt }) => {
-        runAdapterEvent("adapter.audio", async () => {
+        runAudioSerial(async () => {
+          if (activeAdapterGeneration !== generation) {
+            return;
+          }
           await sendAssistantAudio(chunk, {
             format,
             receivedAt
@@ -7350,7 +7376,7 @@ var createVoiceSession = (options) => {
       });
       ttsSession = openedSession;
       openedSession.on("audio", ({ chunk, format, receivedAt }) => {
-        runSerial("tts.audio", async () => {
+        runAudioSerial(async () => {
           if (ttsSession !== openedSession) {
             return;
           }
@@ -7478,6 +7504,7 @@ var createVoiceSession = (options) => {
         try {
           await ttsSession2.send(text);
           charsSent += text.length;
+          lastTtsSendAt = Date.now();
         } catch (error) {
           logger.warn("voice assistant audio send failed", {
             error: toError(error).message,
@@ -7760,6 +7787,7 @@ var createVoiceSession = (options) => {
             turnId: turn.id
           });
           await activeTTSSession.send(output.assistantText);
+          lastTtsSendAt = Date.now();
           if (options.costAccountant) {
             options.costAccountant.recordTTS({
               characters: output.assistantText.length
@@ -8184,10 +8212,12 @@ var createVoiceSession = (options) => {
         if (greetingTTSSession) {
           activeTTSTurnId = greetingTurnId;
           await greetingTTSSession.send(greetingText);
+          lastTtsSendAt = Date.now();
         } else if (options.realtime) {
           const greetingRealtimeSession = await ensureAdapter();
           activeTTSTurnId = greetingTurnId;
           await greetingRealtimeSession.send(greetingText);
+          lastTtsSendAt = Date.now();
         }
       } catch {}
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@absolutejs/voice",
-  "version": "0.0.22-beta.580",
+  "version": "0.0.22-beta.581",
   "description": "Voice primitives and Elysia plugin for AbsoluteJS",
   "repository": {
     "type": "git",