@absolutejs/voice 0.0.22-beta.643 → 0.0.22-beta.644

package/dist/index.js

@@ -4574,6 +4574,8 @@ var createVoiceSession = (options) => {
     user: 0
   };
   const recordingFormats = {};
+  const recordingStartedAt = Date.now();
+  let assistantRecordingEndsAtMs = 0;
   let recordingPersisted = false;
   const captureRecordingChunk = (channel, bytes, format) => {
     if (!recordingConfig || recordingPersisted) {
@@ -4589,11 +4591,33 @@ var createVoiceSession = (options) => {
     if (currentTotal >= recordingMaxBytes) {
       return;
     }
-    const remaining = recordingMaxBytes - currentTotal;
+    const bytesPerSample = 2;
+    const bytesPerSecond = format.sampleRateHz * format.channels * bytesPerSample;
+    let chunkStartMs = currentTotal / bytesPerSecond * 1000;
+    if (channel === "assistant" || currentTotal === 0) {
+      const wallClockOffsetMs = Math.max(0, Date.now() - recordingStartedAt);
+      chunkStartMs = channel === "assistant" ? Math.max(wallClockOffsetMs, assistantRecordingEndsAtMs) : wallClockOffsetMs;
+      const targetByteOffset = Math.floor(chunkStartMs * bytesPerSecond / 1000 / bytesPerSample) * bytesPerSample;
+      const silenceByteLength = Math.min(Math.max(0, targetByteOffset - currentTotal), recordingMaxBytes - currentTotal);
+      if (silenceByteLength > 0) {
+        recordingBuffers[channel].push(new Uint8Array(silenceByteLength));
+        recordingByteTotals[channel] += silenceByteLength;
+      }
+      if (channel === "assistant") {
+        assistantRecordingEndsAtMs = chunkStartMs + bytes.byteLength / bytesPerSecond * 1000;
+      }
+    }
+    const updatedTotal = recordingByteTotals[channel];
+    if (updatedTotal >= recordingMaxBytes) {
+      recordingFormats[channel] = format;
+      return;
+    }
+    const remaining = recordingMaxBytes - updatedTotal;
     const slice = bytes.byteLength <= remaining ? bytes : bytes.subarray(0, remaining);
     recordingBuffers[channel].push(new Uint8Array(slice));
     recordingByteTotals[channel] += slice.byteLength;
     recordingFormats[channel] = format;
+    return { byteLength: slice.byteLength, startMs: chunkStartMs };
   };
   const pruneTurnAudio = () => {
     const replayWindowMs = sttFallback?.replayWindowMs ?? DEFAULT_FALLBACK_REPLAY_MS;
@@ -4921,7 +4945,7 @@ var createVoiceSession = (options) => {
   };
   const sendAssistantAudio = async (chunk, input) => {
     const normalizedChunk = chunk instanceof Uint8Array ? new Uint8Array(chunk) : chunk instanceof ArrayBuffer ? new Uint8Array(chunk.slice(0)) : new Uint8Array(chunk.buffer.slice(chunk.byteOffset, chunk.byteOffset + chunk.byteLength));
-    captureRecordingChunk("assistant", normalizedChunk, input.format);
+    const recordingChunk = captureRecordingChunk("assistant", normalizedChunk, input.format);
     kickCallSilenceWatchdog();
     await send({
       chunkBase64: encodeBase64(normalizedChunk),
@@ -4940,6 +4964,10 @@ var createVoiceSession = (options) => {
     if (activeTTSTurnId) {
       await appendTurnLatencyStage({
         at: input.receivedAt,
+        metadata: recordingChunk ? {
+          recordingByteLength: recordingChunk.byteLength,
+          recordingStartMs: recordingChunk.startMs
+        } : undefined,
         stage: "assistant_audio_received",
         turnId: activeTTSTurnId
       });

package/dist/testing/index.js

@@ -7211,6 +7211,8 @@ var createVoiceSession = (options) => {
     user: 0
   };
   const recordingFormats = {};
+  const recordingStartedAt = Date.now();
+  let assistantRecordingEndsAtMs = 0;
   let recordingPersisted = false;
   const captureRecordingChunk = (channel, bytes2, format) => {
     if (!recordingConfig || recordingPersisted) {
@@ -7226,11 +7228,33 @@ var createVoiceSession = (options) => {
     if (currentTotal >= recordingMaxBytes) {
       return;
     }
-    const remaining = recordingMaxBytes - currentTotal;
+    const bytesPerSample = 2;
+    const bytesPerSecond = format.sampleRateHz * format.channels * bytesPerSample;
+    let chunkStartMs = currentTotal / bytesPerSecond * 1000;
+    if (channel === "assistant" || currentTotal === 0) {
+      const wallClockOffsetMs = Math.max(0, Date.now() - recordingStartedAt);
+      chunkStartMs = channel === "assistant" ? Math.max(wallClockOffsetMs, assistantRecordingEndsAtMs) : wallClockOffsetMs;
+      const targetByteOffset = Math.floor(chunkStartMs * bytesPerSecond / 1000 / bytesPerSample) * bytesPerSample;
+      const silenceByteLength = Math.min(Math.max(0, targetByteOffset - currentTotal), recordingMaxBytes - currentTotal);
+      if (silenceByteLength > 0) {
+        recordingBuffers[channel].push(new Uint8Array(silenceByteLength));
+        recordingByteTotals[channel] += silenceByteLength;
+      }
+      if (channel === "assistant") {
+        assistantRecordingEndsAtMs = chunkStartMs + bytes2.byteLength / bytesPerSecond * 1000;
+      }
+    }
+    const updatedTotal = recordingByteTotals[channel];
+    if (updatedTotal >= recordingMaxBytes) {
+      recordingFormats[channel] = format;
+      return;
+    }
+    const remaining = recordingMaxBytes - updatedTotal;
     const slice = bytes2.byteLength <= remaining ? bytes2 : bytes2.subarray(0, remaining);
     recordingBuffers[channel].push(new Uint8Array(slice));
     recordingByteTotals[channel] += slice.byteLength;
     recordingFormats[channel] = format;
+    return { byteLength: slice.byteLength, startMs: chunkStartMs };
   };
   const pruneTurnAudio = () => {
     const replayWindowMs = sttFallback?.replayWindowMs ?? DEFAULT_FALLBACK_REPLAY_MS;
@@ -7558,7 +7582,7 @@ var createVoiceSession = (options) => {
   };
   const sendAssistantAudio = async (chunk, input) => {
     const normalizedChunk = chunk instanceof Uint8Array ? new Uint8Array(chunk) : chunk instanceof ArrayBuffer ? new Uint8Array(chunk.slice(0)) : new Uint8Array(chunk.buffer.slice(chunk.byteOffset, chunk.byteOffset + chunk.byteLength));
-    captureRecordingChunk("assistant", normalizedChunk, input.format);
+    const recordingChunk = captureRecordingChunk("assistant", normalizedChunk, input.format);
     kickCallSilenceWatchdog();
     await send({
       chunkBase64: encodeBase64(normalizedChunk),
@@ -7577,6 +7601,10 @@ var createVoiceSession = (options) => {
     if (activeTTSTurnId) {
       await appendTurnLatencyStage({
         at: input.receivedAt,
+        metadata: recordingChunk ? {
+          recordingByteLength: recordingChunk.byteLength,
+          recordingStartMs: recordingChunk.startMs
+        } : undefined,
         stage: "assistant_audio_received",
         turnId: activeTTSTurnId
       });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@absolutejs/voice",
-  "version": "0.0.22-beta.643",
+  "version": "0.0.22-beta.644",
   "description": "Voice primitives and Elysia plugin for AbsoluteJS",
   "repository": {
     "type": "git",