@jskit-ai/assistant-voice 0.1.9 → 0.1.12

package/README.md

@@ -82,13 +82,18 @@ The host's `avatar` slot receives state, mouthLevel, mouthPose, avatar and bindi
 
 Opening a target does not start capture. Tap Talk for hands-free, or hold it for
 push-to-talk and release to send; connection setup is automatic.
-Pause finishes the current utterance, flushing the audio tail before muting.
+Pause stops new microphone input immediately and retains its buffered audio for
+the original final flush before recognition finishes the current utterance.
 Completed words follow the captured review policy. Pending delivery does not disable
 the microphone: tap can resume the existing continuous capture or start another
 one while preserving the prior pending transcript. A busy Pause or hold release
-mutes newer capture and finishes it through the original owner after that prior
-admission settles. Resuming first keeps the capture open. No pending words are
+mutes newer capture without discarding its buffered tail and finishes it through
+the original owner after that prior admission settles. Resuming first keeps the
+capture and its retained tail open. No pending words are
 replaced or appended. The icon reflects actual unmuted capture, not readiness.
+When the daemon rejects a stale endpoint commit, its exact turn and revision
+release only that commit. A waiting Pause can then finalize the same recording;
+an obsolete acknowledgement cannot release a newer commit or submit provisional words.
 Automatic sends keep the call layout fixed. A failed send exposes the retained
 transcript for review and retry with the same message identity. An application
 can call `beginTranscriptEdit(messageId)` for its exact unsent preview; this

package/examples/conversation/package.json

@@ -9,14 +9,14 @@
   },
   "dependencies": {
     "@fastify/websocket": "^11.2.0",
-    "@jskit-ai/assistant-core": "0.1.209",
-    "@jskit-ai/assistant-runtime": "0.1.202",
-    "@jskit-ai/assistant-voice": "0.1.9",
-    "@jskit-ai/connectors-catalog": "0.1.48",
-    "@jskit-ai/http-runtime": "0.1.229",
-    "@jskit-ai/kernel": "0.1.231",
-    "@jskit-ai/realtime": "0.1.228",
-    "@jskit-ai/shell-web": "0.1.235",
+    "@jskit-ai/assistant-core": "0.1.212",
+    "@jskit-ai/assistant-runtime": "0.1.205",
+    "@jskit-ai/assistant-voice": "0.1.12",
+    "@jskit-ai/connectors-catalog": "0.1.51",
+    "@jskit-ai/http-runtime": "0.1.232",
+    "@jskit-ai/kernel": "0.1.234",
+    "@jskit-ai/realtime": "0.1.231",
+    "@jskit-ai/shell-web": "0.1.238",
     "@mdi/js": "^7.4.47",
     "@tanstack/vue-query": "^5.90.5",
     "fastify": "^5.8.0",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jskit-ai/assistant-voice",
-  "version": "0.1.9",
+  "version": "0.1.12",
   "type": "module",
   "description": "Duplex voice conversations with application-owned targets, authorization and speech services.",
   "scripts": {
@@ -18,7 +18,7 @@
     "examples"
   ],
   "dependencies": {
-    "@jskit-ai/assistant-core": "0.1.209",
+    "@jskit-ai/assistant-core": "0.1.212",
     "@mdi/js": "^7.4.47",
     "ws": "^8.20.1"
   },

package/src/client/audioCapture.js

@@ -12,8 +12,10 @@ class AssistantVoiceCapture extends AudioWorkletProcessor {
     this.port.onmessage = (event) => {
       if (event.data?.type === "mute") {
         this.muted = event.data.muted;
-        this.frames = [];
-        this.length = 0;
+        if (!event.data.preserveBuffered) {
+          this.frames = [];
+          this.length = 0;
+        }
         return;
       }
       if (event.data?.type !== "flush") return;
@@ -108,12 +110,13 @@ async function createMicrophoneCapture({ onLevel, onPcm, targetSampleRate = 16_0
     silence.gain.value = 0;
     let finishFlush = null;
     let muted = false;
+    let preserveBuffered = false;
     capture.port.onmessage = (event) => {
       if (event.data?.type === "flushed") {
         finishFlush?.();
         return;
       }
-      if (muted) return;
+      if (muted && !preserveBuffered) return;
       const samples = event.data instanceof Float32Array
         ? event.data
         : new Float32Array(event.data);
@@ -154,11 +157,14 @@ async function createMicrophoneCapture({ onLevel, onPcm, targetSampleRate = 16_0
       }
     });
     return Object.freeze({
-      setMuted(value) {
-        if (muted === Boolean(value)) return;
+      setMuted(value, { preserveBuffered: retainTail = false } = {}) {
+        const nextPreserveBuffered = Boolean(value) && retainTail;
+        if (muted === Boolean(value) && preserveBuffered === nextPreserveBuffered) return;
+        const keepBuffered = nextPreserveBuffered || (!value && preserveBuffered);
         muted = Boolean(value);
+        preserveBuffered = nextPreserveBuffered;
         for (const track of stream.getTracks()) track.enabled = !muted;
-        capture.port.postMessage({ type: "mute", muted });
+        capture.port.postMessage({ type: "mute", muted, ...(keepBuffered ? { preserveBuffered: true } : {}) });
         if (muted) onLevel?.(0);
       },
       close() {

package/src/client/voiceConversation.js

@@ -409,10 +409,10 @@ export function useVoiceConversation(binding, { socketUrl, createTransport = use
       if (recording === next) { next.started = started; if (!started) recording = null; }
     } catch (cause) { if (recording === next) { recording = null; error.value = cause.message; } }
   }
-  async function toggleMicrophoneMuted() {
+  async function toggleMicrophoneMuted({ preserveBuffered = false } = {}) {
     // A person's explicit microphone choice supersedes an Edit-owned pause.
     if (pendingTranscript.value) delete pendingTranscript.value.editPausedCaptureId;
-    voice.setMicrophoneMuted(!microphoneMuted.value);
+    voice.setMicrophoneMuted(!microphoneMuted.value, { preserveBuffered });
     if (!microphoneMuted.value && recording) delete recording.finishAfterPending;
     clearTimeout(stopTimer);
     if (microphoneMuted.value && starting.value) await cancelRecording();
@@ -455,7 +455,7 @@ export function useVoiceConversation(binding, { socketUrl, createTransport = use
     try {
       if (voice.listening.value && (pendingTranscript.value || sending.value || committing)) {
         if (recording) recording.finishAfterPending = true;
-        voice.setMicrophoneMuted(true);
+        voice.setMicrophoneMuted(true, { preserveBuffered: true });
         return;
       }
       if (voice.listening.value) await voice.stopListening();
@@ -474,7 +474,7 @@ export function useVoiceConversation(binding, { socketUrl, createTransport = use
       // Pause must finish the newer recording after the existing admission,
       // without replacing the one pending transcript or stopping output.
       if (!microphoneMuted.value && recording) recording.finishAfterPending = true;
-      await toggleMicrophoneMuted();
+      await toggleMicrophoneMuted({ preserveBuffered: true });
       return;
     }
     if (committing) return;
@@ -482,10 +482,10 @@ export function useVoiceConversation(binding, { socketUrl, createTransport = use
     if (capturing.value && !voice.listening.value) return;
     if (live.value && callMode.value === "hands-free") {
       if (!microphoneMuted.value && voice.listening.value) {
-        // Pause is an explicit utterance boundary. Flush before muting so the
-        // final sound is included; resuming starts a fresh recording.
+        // Stop new input immediately, retaining the captured tail for the
+        // original Stop drain. Resuming starts a fresh recording.
+        voice.setMicrophoneMuted(true, { preserveBuffered: true });
         await voice.stopListening();
-        voice.setMicrophoneMuted(true);
       } else if (microphoneMuted.value) await toggleMicrophoneMuted();
       else await startRecording(false, true);
       return;
@@ -608,7 +608,11 @@ export function useVoiceConversation(binding, { socketUrl, createTransport = use
       if (live.value && !microphoneMuted.value && voice.partialTranscript.value === text) stopSpeech();
     }, 120);
   });
-  watch([pendingTranscript, sending, voice.utteranceReset], () => {
+  watch([pendingTranscript, sending, voice.utteranceReset, voice.utteranceStale], () => {
+    const stale = voice.utteranceStale.value;
+    if (committing && stale?.turnId === committing.turnId && stale.revision === committing.revision) {
+      committing = null;
+    }
     if (!disposed && live.value && !pushHolding.value && !changingCallMode.value &&
         !pendingTranscript.value && !sending.value && !committing && microphoneMuted.value &&
         recording?.finishAfterPending && voice.listening.value) {

package/src/client/voiceTransport.js

@@ -41,6 +41,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
   const endpoint = ref(null);
   const completedUtterance = ref(null);
   const utteranceReset = ref(null);
+  const utteranceStale = ref(null);
   const partialTranscript = ref("");
   const muted = ref(false);
   const microphoneMuted = ref(false);
@@ -63,6 +64,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
   let lastPongAt = 0;
   let lastSpeechProgressAt = 0;
   let microphone = null;
+  let preserveBufferedMicrophone = false;
   const openingMicrophones = new Set();
   let microphoneClaim = null;
   const speechClaim = Symbol("voice playback");
@@ -444,6 +446,8 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
         utteranceReset.value = { turnId: message.turnId, revision: message.revision };
         transcript.value = "";
         partialTranscript.value = "";
+      } else {
+        utteranceStale.value = { turnId: message.turnId, revision: message.revision };
       }
       return;
     }
@@ -629,6 +633,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
     }
     microphoneClaim = Symbol("voice microphone");
     microphoneOwner = microphoneClaim;
+    preserveBufferedMicrophone = false;
     const turnId = voiceTurnId("listen");
     pendingListenTurnId = turnId;
     if (interruptSpeechOnListen) stopSpeaking();
@@ -647,7 +652,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
             }
           },
           onPcm(frame) {
-            if (!microphoneMuted.value && socket?.readyState === WEBSOCKET_OPEN && activeListenTurnId.value === turnId) {
+            if ((!microphoneMuted.value || preserveBufferedMicrophone) && socket?.readyState === WEBSOCKET_OPEN && activeListenTurnId.value === turnId) {
               if (socket.bufferedAmount > 32 * 1024) {
                 connectionLost("The microphone connection fell behind. Review the unfinished recording before sending.");
                 return;
@@ -843,9 +848,10 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
     muted.value = !muted.value;
   }
 
-  function setMicrophoneMuted(value) {
+  function setMicrophoneMuted(value, { preserveBuffered = false } = {}) {
     microphoneMuted.value = Boolean(value);
-    microphone?.setMuted(microphoneMuted.value);
+    preserveBufferedMicrophone = microphoneMuted.value && preserveBuffered;
+    microphone?.setMuted(microphoneMuted.value, { preserveBuffered });
     if (microphoneMuted.value) inputLevel.value = 0;
   }
 
@@ -900,6 +906,7 @@ function useVoiceTransport({ socketUrl, speechEnabled = true, interruptSpeechOnL
     captureState,
     completedUtterance,
     utteranceReset,
+    utteranceStale,
     endpoint,
     finishUtterance,
     finishCurrentPhrase,