@absolutejs/voice 0.0.22-beta.573 → 0.0.22-beta.575

package/dist/client/audioPlayer.d.ts

@@ -8,6 +8,9 @@ type MinimalAudioBufferSourceNode = {
     connect: (destination: unknown) => void;
     disconnect?: () => void;
     onended: (() => void) | null;
+    playbackRate?: {
+        value: number;
+    };
     start: (when?: number) => void;
     stop?: () => void;
 };

package/dist/client/htmxBootstrap.js

@@ -1536,6 +1536,9 @@ var createVoiceController = (path, options = {}) => {
 // src/client/audioPlayer.ts
 var DEFAULT_LOOKAHEAD_MS = 15;
 var DEFAULT_VOLUME = 1;
+var DEFAULT_PLAYBACK_RATE = 1;
+var MIN_PLAYBACK_RATE = 0.5;
+var MAX_PLAYBACK_RATE = 2;
 var createInitialState3 = () => ({
   activeSourceCount: 0,
   error: null,
@@ -1558,6 +1561,12 @@ var clampVolume = (volume) => {
   }
   return Math.min(1, Math.max(0, volume));
 };
+var clampPlaybackRate = (rate) => {
+  if (typeof rate !== "number" || !Number.isFinite(rate)) {
+    return DEFAULT_PLAYBACK_RATE;
+  }
+  return Math.min(MAX_PLAYBACK_RATE, Math.max(MIN_PLAYBACK_RATE, rate));
+};
 var decodePCM16LEChunk = (audioContext, chunk) => {
   const { format } = chunk;
   if (format.container !== "raw" || format.encoding !== "pcm_s16le") {
@@ -1591,6 +1600,7 @@ var createVoiceAudioPlayer = (source, options = {}) => {
   let audioContext = null;
   let outputNode = null;
   let volume = clampVolume(options.volume);
+  let playbackRate = clampPlaybackRate(options.playbackRate);
   let queueEndTime = 0;
   let syncPromise = Promise.resolve();
   let interruptStartedAt = null;
@@ -1692,6 +1702,9 @@ var createVoiceAudioPlayer = (source, options = {}) => {
     const buffer = decodePCM16LEChunk(context, chunk);
     const node = context.createBufferSource();
     node.buffer = buffer;
+    if (node.playbackRate) {
+      node.playbackRate.value = playbackRate;
+    }
     node.connect(outputNode ?? context.destination);
     node.onended = () => {
       sourceNodes.delete(node);
@@ -1703,7 +1716,7 @@ var createVoiceAudioPlayer = (source, options = {}) => {
       maybeResolveInterrupt();
     };
     const startAt = Math.max(context.currentTime + lookaheadSeconds, queueEndTime);
-    queueEndTime = startAt + buffer.duration;
+    queueEndTime = startAt + buffer.duration / playbackRate;
     sourceNodes.add(node);
     setState({
       activeSourceCount: sourceNodes.size,
@@ -1848,12 +1861,18 @@ var createVoiceAudioPlayer = (source, options = {}) => {
         isPlaying: false
       });
     },
+    get playbackRate() {
+      return playbackRate;
+    },
     get processedChunkCount() {
       return state.processedChunkCount;
     },
     get queuedChunkCount() {
       return state.queuedChunkCount;
     },
+    setPlaybackRate: (nextRate) => {
+      playbackRate = clampPlaybackRate(nextRate);
+    },
     setVolume: (nextVolume) => {
       volume = clampVolume(nextVolume);
       applyOutputGain(audioContext);

package/dist/client/index.js

@@ -373,6 +373,9 @@ var createVoiceConnection = (path, options = {}) => {
 // src/client/audioPlayer.ts
 var DEFAULT_LOOKAHEAD_MS = 15;
 var DEFAULT_VOLUME = 1;
+var DEFAULT_PLAYBACK_RATE = 1;
+var MIN_PLAYBACK_RATE = 0.5;
+var MAX_PLAYBACK_RATE = 2;
 var createInitialState = () => ({
   activeSourceCount: 0,
   error: null,
@@ -395,6 +398,12 @@ var clampVolume = (volume) => {
   }
   return Math.min(1, Math.max(0, volume));
 };
+var clampPlaybackRate = (rate) => {
+  if (typeof rate !== "number" || !Number.isFinite(rate)) {
+    return DEFAULT_PLAYBACK_RATE;
+  }
+  return Math.min(MAX_PLAYBACK_RATE, Math.max(MIN_PLAYBACK_RATE, rate));
+};
 var decodePCM16LEChunk = (audioContext, chunk) => {
   const { format } = chunk;
   if (format.container !== "raw" || format.encoding !== "pcm_s16le") {
@@ -428,6 +437,7 @@ var createVoiceAudioPlayer = (source, options = {}) => {
   let audioContext = null;
   let outputNode = null;
   let volume = clampVolume(options.volume);
+  let playbackRate = clampPlaybackRate(options.playbackRate);
   let queueEndTime = 0;
   let syncPromise = Promise.resolve();
   let interruptStartedAt = null;
@@ -529,6 +539,9 @@ var createVoiceAudioPlayer = (source, options = {}) => {
     const buffer = decodePCM16LEChunk(context, chunk);
     const node = context.createBufferSource();
     node.buffer = buffer;
+    if (node.playbackRate) {
+      node.playbackRate.value = playbackRate;
+    }
     node.connect(outputNode ?? context.destination);
     node.onended = () => {
       sourceNodes.delete(node);
@@ -540,7 +553,7 @@ var createVoiceAudioPlayer = (source, options = {}) => {
       maybeResolveInterrupt();
     };
     const startAt = Math.max(context.currentTime + lookaheadSeconds, queueEndTime);
-    queueEndTime = startAt + buffer.duration;
+    queueEndTime = startAt + buffer.duration / playbackRate;
     sourceNodes.add(node);
     setState({
       activeSourceCount: sourceNodes.size,
@@ -685,12 +698,18 @@ var createVoiceAudioPlayer = (source, options = {}) => {
         isPlaying: false
       });
     },
+    get playbackRate() {
+      return playbackRate;
+    },
     get processedChunkCount() {
       return state.processedChunkCount;
     },
     get queuedChunkCount() {
       return state.queuedChunkCount;
     },
+    setPlaybackRate: (nextRate) => {
+      playbackRate = clampPlaybackRate(nextRate);
+    },
     setVolume: (nextVolume) => {
       volume = clampVolume(nextVolume);
       applyOutputGain(audioContext);

package/dist/core/types.d.ts

@@ -1183,6 +1183,14 @@ export type VoiceAudioPlayerOptions = {
     autoStart?: boolean;
     createAudioContext?: () => AudioContext;
     lookaheadMs?: number;
+    /**
+     * Playback speed multiplier for the assistant's speech. 1 = normal. Clamped
+     * to [0.5, 2]. Pitch shifts with the rate (Web Audio playbackRate), so keep
+     * UI ranges modest (≈0.85–1.25) to stay natural. Can be changed live via
+     * setPlaybackRate — already-scheduled chunks keep their rate; new chunks
+     * adopt the new one.
+     */
+    playbackRate?: number;
     volume?: number;
 };
 export type VoiceDuplexControllerOptions = VoiceControllerOptions & {
@@ -1313,8 +1321,10 @@ export type VoiceAudioPlayer = {
     lastInterruptLatencyMs?: number;
     lastPlaybackStopLatencyMs?: number;
     pause: () => Promise<void>;
+    playbackRate: number;
     processedChunkCount: number;
     queuedChunkCount: number;
+    setPlaybackRate: (rate: number) => void;
     setVolume: (volume: number) => void;
     start: () => Promise<void>;
     subscribe: (subscriber: () => void) => () => void;

package/dist/index.js

@@ -3870,6 +3870,7 @@ var createVoiceSession = (options) => {
   let adapterGenerationCounter = 0;
   let activeAdapterGeneration = 0;
   let activeTTSTurnId;
+  let assistantSpeechEndsAt = 0;
   let fillerTimer = null;
   let fillerActive = false;
   let fillerToken = 0;
@@ -4262,6 +4263,7 @@ var createVoiceSession = (options) => {
       return;
     }
     activeTTSTurnId = undefined;
+    assistantSpeechEndsAt = Date.now();
     appendTurnLatencyStage({
       metadata: { reason },
       stage: "tts_canceled",
@@ -4304,6 +4306,12 @@ var createVoiceSession = (options) => {
       turnId: activeTTSTurnId,
       type: "audio"
     });
+    const bytesPerSample = input.format.encoding === "pcm_s16le" ? 2 : 1;
+    const bytesPerSecond = input.format.sampleRateHz * input.format.channels * bytesPerSample;
+    if (bytesPerSecond > 0) {
+      const chunkMs = normalizedChunk.byteLength / bytesPerSecond * 1000;
+      assistantSpeechEndsAt = Math.max(assistantSpeechEndsAt, Date.now()) + chunkMs;
+    }
     if (activeTTSTurnId) {
       await appendTurnLatencyStage({
         at: input.receivedAt,
@@ -4413,6 +4421,20 @@ var createVoiceSession = (options) => {
       session
     });
   };
+  const DRAIN_POLL_MS = 200;
+  const DRAIN_TAIL_BUFFER_MS = 300;
+  const DRAIN_MAX_MS = 12000;
+  const drainAssistantSpeech = async () => {
+    const startedAt = Date.now();
+    while (Date.now() - startedAt < DRAIN_MAX_MS) {
+      const remaining = assistantSpeechEndsAt + DRAIN_TAIL_BUFFER_MS - Date.now();
+      if (remaining <= 0)
+        return;
+      await new Promise((resolve) => {
+        setTimeout(resolve, Math.min(remaining, DRAIN_POLL_MS));
+      });
+    }
+  };
   const completeInternal = async (result, input = {}) => {
     clearSilenceTimer();
     const disposition = input.disposition ?? "completed";
@@ -4446,6 +4468,9 @@ var createVoiceSession = (options) => {
     if (!didComplete) {
       return;
     }
+    if (disposition === "completed") {
+      await drainAssistantSpeech();
+    }
     await appendTrace({
       payload: {
         disposition,

package/dist/testing/index.js

@@ -1580,6 +1580,9 @@ var buildSessionCorrectionAudit = (raw, generic, experimental, benchmarkSeeded,
 // src/client/audioPlayer.ts
 var DEFAULT_LOOKAHEAD_MS = 15;
 var DEFAULT_VOLUME = 1;
+var DEFAULT_PLAYBACK_RATE = 1;
+var MIN_PLAYBACK_RATE = 0.5;
+var MAX_PLAYBACK_RATE = 2;
 var createInitialState = () => ({
   activeSourceCount: 0,
   error: null,
@@ -1602,6 +1605,12 @@ var clampVolume = (volume) => {
   }
   return Math.min(1, Math.max(0, volume));
 };
+var clampPlaybackRate = (rate) => {
+  if (typeof rate !== "number" || !Number.isFinite(rate)) {
+    return DEFAULT_PLAYBACK_RATE;
+  }
+  return Math.min(MAX_PLAYBACK_RATE, Math.max(MIN_PLAYBACK_RATE, rate));
+};
 var decodePCM16LEChunk = (audioContext, chunk) => {
   const { format } = chunk;
   if (format.container !== "raw" || format.encoding !== "pcm_s16le") {
@@ -1635,6 +1644,7 @@ var createVoiceAudioPlayer = (source, options = {}) => {
   let audioContext = null;
   let outputNode = null;
   let volume = clampVolume(options.volume);
+  let playbackRate = clampPlaybackRate(options.playbackRate);
   let queueEndTime = 0;
   let syncPromise = Promise.resolve();
   let interruptStartedAt = null;
@@ -1736,6 +1746,9 @@ var createVoiceAudioPlayer = (source, options = {}) => {
     const buffer = decodePCM16LEChunk(context, chunk);
     const node = context.createBufferSource();
     node.buffer = buffer;
+    if (node.playbackRate) {
+      node.playbackRate.value = playbackRate;
+    }
     node.connect(outputNode ?? context.destination);
     node.onended = () => {
       sourceNodes.delete(node);
@@ -1747,7 +1760,7 @@ var createVoiceAudioPlayer = (source, options = {}) => {
       maybeResolveInterrupt();
     };
     const startAt = Math.max(context.currentTime + lookaheadSeconds, queueEndTime);
-    queueEndTime = startAt + buffer.duration;
+    queueEndTime = startAt + buffer.duration / playbackRate;
     sourceNodes.add(node);
     setState({
       activeSourceCount: sourceNodes.size,
@@ -1892,12 +1905,18 @@ var createVoiceAudioPlayer = (source, options = {}) => {
         isPlaying: false
       });
     },
+    get playbackRate() {
+      return playbackRate;
+    },
     get processedChunkCount() {
       return state.processedChunkCount;
     },
     get queuedChunkCount() {
       return state.queuedChunkCount;
     },
+    setPlaybackRate: (nextRate) => {
+      playbackRate = clampPlaybackRate(nextRate);
+    },
     setVolume: (nextVolume) => {
       volume = clampVolume(nextVolume);
       applyOutputGain(audioContext);
@@ -5786,6 +5805,7 @@ var createVoiceSession = (options) => {
   let adapterGenerationCounter = 0;
   let activeAdapterGeneration = 0;
   let activeTTSTurnId;
+  let assistantSpeechEndsAt = 0;
   let fillerTimer = null;
   let fillerActive = false;
   let fillerToken = 0;
@@ -6178,6 +6198,7 @@ var createVoiceSession = (options) => {
       return;
     }
     activeTTSTurnId = undefined;
+    assistantSpeechEndsAt = Date.now();
     appendTurnLatencyStage({
       metadata: { reason },
       stage: "tts_canceled",
@@ -6220,6 +6241,12 @@ var createVoiceSession = (options) => {
       turnId: activeTTSTurnId,
       type: "audio"
     });
+    const bytesPerSample = input.format.encoding === "pcm_s16le" ? 2 : 1;
+    const bytesPerSecond = input.format.sampleRateHz * input.format.channels * bytesPerSample;
+    if (bytesPerSecond > 0) {
+      const chunkMs = normalizedChunk.byteLength / bytesPerSecond * 1000;
+      assistantSpeechEndsAt = Math.max(assistantSpeechEndsAt, Date.now()) + chunkMs;
+    }
     if (activeTTSTurnId) {
       await appendTurnLatencyStage({
         at: input.receivedAt,
@@ -6329,6 +6356,20 @@ var createVoiceSession = (options) => {
       session
     });
   };
+  const DRAIN_POLL_MS = 200;
+  const DRAIN_TAIL_BUFFER_MS = 300;
+  const DRAIN_MAX_MS = 12000;
+  const drainAssistantSpeech = async () => {
+    const startedAt = Date.now();
+    while (Date.now() - startedAt < DRAIN_MAX_MS) {
+      const remaining = assistantSpeechEndsAt + DRAIN_TAIL_BUFFER_MS - Date.now();
+      if (remaining <= 0)
+        return;
+      await new Promise((resolve2) => {
+        setTimeout(resolve2, Math.min(remaining, DRAIN_POLL_MS));
+      });
+    }
+  };
   const completeInternal = async (result, input = {}) => {
     clearSilenceTimer();
     const disposition = input.disposition ?? "completed";
@@ -6362,6 +6403,9 @@ var createVoiceSession = (options) => {
     if (!didComplete) {
       return;
     }
+    if (disposition === "completed") {
+      await drainAssistantSpeech();
+    }
     await appendTrace({
       payload: {
         disposition,

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@absolutejs/voice",
-	"version": "0.0.22-beta.573",
+	"version": "0.0.22-beta.575",
 	"description": "Voice primitives and Elysia plugin for AbsoluteJS",
 	"repository": {
 		"type": "git",