@absolutejs/voice 0.0.22-beta.642 → 0.0.22-beta.644

package/dist/core/hardenedFetch.d.ts

@@ -1,3 +1 @@
-export declare const hardenFetch: (baseFetch?: typeof fetch) => ((input: Parameters<typeof fetch>[0], init: Parameters<typeof fetch>[1]) => Promise<Response>) & {
-    preconnect: typeof fetch.preconnect;
-};
+export declare const hardenFetch: (baseFetch?: typeof fetch) => (input: Parameters<typeof fetch>[0], init: Parameters<typeof fetch>[1]) => Promise<Response>;

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
       });
@@ -41820,16 +41848,21 @@ var oneAttempt = async (baseFetch, input, init) => {
     callerSignal?.removeEventListener("abort", onCallerAbort);
   }
 };
-var hardenFetch = (baseFetch = globalThis.fetch) => Object.assign(async (input, init) => {
-  try {
-    return await oneAttempt(baseFetch, input, init);
-  } catch (error) {
-    if (init?.signal?.aborted)
-      throw error;
-    console.warn(`[voice] hardened fetch retrying on a fresh connection: ${error instanceof Error ? error.message : String(error)}`);
-    return oneAttempt(baseFetch, input, init);
-  }
-}, { preconnect: baseFetch.preconnect.bind(baseFetch) });
+var hardenFetch = (baseFetch = globalThis.fetch) => {
+  const hardenedFetch = async (input, init) => {
+    try {
+      return await oneAttempt(baseFetch, input, init);
+    } catch (error) {
+      if (init?.signal?.aborted)
+        throw error;
+      console.warn(`[voice] hardened fetch retrying on a fresh connection: ${error instanceof Error ? error.message : String(error)}`);
+      return oneAttempt(baseFetch, input, init);
+    }
+  };
+  return typeof baseFetch.preconnect === "function" ? Object.assign(hardenedFetch, {
+    preconnect: baseFetch.preconnect.bind(baseFetch)
+  }) : hardenedFetch;
+};
 // src/core/mcpToolset.ts
 var flattenContent = (result) => {
   const blocks = result.content ?? [];

package/dist/testing/index.js

@@ -328,12 +328,7 @@ 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
@@ -4759,16 +4754,21 @@ var oneAttempt = async (baseFetch, input, init) => {
     callerSignal?.removeEventListener("abort", onCallerAbort);
   }
 };
-var hardenFetch = (baseFetch = globalThis.fetch) => Object.assign(async (input, init) => {
-  try {
-    return await oneAttempt(baseFetch, input, init);
-  } catch (error) {
-    if (init?.signal?.aborted)
-      throw error;
-    console.warn(`[voice] hardened fetch retrying on a fresh connection: ${error instanceof Error ? error.message : String(error)}`);
-    return oneAttempt(baseFetch, input, init);
-  }
-}, { preconnect: baseFetch.preconnect.bind(baseFetch) });
+var hardenFetch = (baseFetch = globalThis.fetch) => {
+  const hardenedFetch = async (input, init) => {
+    try {
+      return await oneAttempt(baseFetch, input, init);
+    } catch (error) {
+      if (init?.signal?.aborted)
+        throw error;
+      console.warn(`[voice] hardened fetch retrying on a fresh connection: ${error instanceof Error ? error.message : String(error)}`);
+      return oneAttempt(baseFetch, input, init);
+    }
+  };
+  return typeof baseFetch.preconnect === "function" ? Object.assign(hardenedFetch, {
+    preconnect: baseFetch.preconnect.bind(baseFetch)
+  }) : hardenedFetch;
+};
 
 // src/core/modelAdapters.ts
 var normalizeVoiceUsage = (provider, model, raw) => {
@@ -6010,15 +6010,9 @@ 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;
@@ -7217,6 +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) {
@@ -7232,11 +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;
@@ -7564,7 +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),
@@ -7583,6 +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.642",
+  "version": "0.0.22-beta.644",
   "description": "Voice primitives and Elysia plugin for AbsoluteJS",
   "repository": {
     "type": "git",
@@ -117,7 +117,7 @@
     "@absolutejs/absolute": ">=0.19.0-beta.646",
     "@absolutejs/ai": ">=0.0.5",
     "@angular/core": ">=21.0.0",
-    "drizzle-orm": ">=1.0.0-rc.1",
+    "drizzle-orm": ">=1.0.0-rc.4 <2",
     "elysia": ">=1.4.18",
     "react": ">=19.0.0",
     "vue": ">=3.5.0"
@@ -154,7 +154,7 @@
     "@types/bun": "1.3.9",
     "@types/react": "19.2.0",
     "@typescript-eslint/parser": "^8.57.2",
-    "drizzle-orm": "1.0.0-rc.3",
+    "drizzle-orm": "1.0.0-rc.4",
     "elysia": "1.4.18",
     "eslint": "^10.1.0",
     "eslint-plugin-absolute": "0.11.0-beta.3",