@absolutejs/voice 0.0.22-beta.574 → 0.0.22-beta.576

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/cachedTTS.d.ts

@@ -1,4 +1,21 @@
-import type { TTSAdapter, TTSAdapterOpenOptions } from "./types";
+import type { TTSAdapter, TTSAdapterOpenOptions, TTSAudioEvent } from "./types";
+/**
+ * Optional persistent backing store for the cache — an L2 behind the in-memory
+ * LRU. Lets rendered audio survive process restarts/deploys so a fixed prompt
+ * (e.g. a greeting) is synthesized once *ever* per content key, not once per
+ * process. The store is content-addressed by the same `keyFor` key, so a
+ * changed prompt/voice/model naturally lands on a new key and re-renders.
+ *
+ * The store is told `TTSAudioEvent[]` and must return the same on read; how it
+ * serializes the binary `chunk`s (base64 in JSON, bytea, a file, etc.) is up to
+ * the implementation. `get` returns `null`/`undefined` on a miss. Both may be
+ * sync or async; errors should be swallowed by the implementation (a store
+ * failure must never break playback — the wrapper falls back to live render).
+ */
+export type CachedTTSStore = {
+    get: (key: string) => Promise<TTSAudioEvent[] | null | undefined> | TTSAudioEvent[] | null | undefined;
+    set: (key: string, events: TTSAudioEvent[]) => Promise<void> | void;
+};
 export type CachedTTSOptions = {
     /**
      * Return a stable cache key for an utterance whose synthesized audio should
@@ -13,8 +30,15 @@ export type CachedTTSOptions = {
      * (and re-caches) while the old entry is simply orphaned.
      */
     keyFor: (text: string, openOptions: TTSAdapterOpenOptions) => string | null | undefined;
-    /** Max distinct utterances to retain (LRU by insertion). Default 32. */
+    /** Max distinct utterances to retain in memory (LRU by insertion). Default 32. */
     maxEntries?: number;
+    /**
+     * Optional persistent L2 store (see {@link CachedTTSStore}). When set, an
+     * in-memory miss consults the store before rendering; a store hit is replayed
+     * and promoted into memory, and a fresh render is written through to it. Omit
+     * for memory-only behaviour (unchanged).
+     */
+    store?: CachedTTSStore;
 };
 /**
  * Wrap a TTS adapter so selected utterances are synthesized once and replayed

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.d.ts

@@ -223,7 +223,7 @@ export type { VoiceSimulationSuiteAssertionInput, VoiceSimulationSuiteAssertionR
 export type { VoiceWorkflowContract, VoiceWorkflowContractDefinition, VoiceWorkflowContractField, VoiceWorkflowContractFieldMatch, VoiceWorkflowContractPresetName, VoiceWorkflowContractPresetOptions, VoiceWorkflowContractTracePayload, VoiceWorkflowContractValidation, VoiceWorkflowContractValidationIssue, VoiceWorkflowOutcome, } from "./core/workflowContract";
 export type { VoiceSessionListHTMLHandlerOptions, VoiceSessionListItem, VoiceSessionListOptions, VoiceSessionListRoutesOptions, VoiceSessionListStatus, VoiceProviderFallbackRecoverySummary, VoiceSessionReplay, VoiceSessionReplayHTMLHandlerOptions, VoiceSessionReplayOptions, VoiceSessionReplayRoutesOptions, VoiceSessionReplayTurn, } from "./core/sessionReplay";
 export type { AnthropicVoiceAssistantModelOptions, GeminiVoiceAssistantModelOptions, OpenAIVoiceAssistantModelOptions, VoiceProviderRouterEvent, VoiceProviderRouterFallbackMode, VoiceProviderRouterHealthOptions, VoiceProviderRouterOptions, VoiceProviderOrchestrationProfile, VoiceProviderOrchestrationProfileOptions, VoiceProviderOrchestrationResolvedSurface, VoiceProviderOrchestrationSurface, VoiceProviderRouterPolicy, VoiceProviderRouterPolicyPreset, VoiceProviderRouterPolicyWeights, VoiceProviderRouterProviderHealth, VoiceProviderRouterProviderProfile, VoiceProviderRouterStrategy, VoiceJSONAssistantModelHandler, VoiceJSONAssistantModelOptions, } from "./core/modelAdapters";
-export type { CachedTTSOptions } from "./core/cachedTTS";
+export type { CachedTTSOptions, CachedTTSStore } from "./core/cachedTTS";
 export type { OpenAIVoiceTTSOptions, OpenAIVoiceTTSVoice, } from "./core/openaiTTS";
 export type { VoiceProviderHealthStatus, VoiceProviderHealthSummary, VoiceProviderHealthSummaryOptions, } from "./core/providerHealth";
 export type { VoiceProviderCapabilityDefinition, VoiceProviderCapabilityHandlerOptions, VoiceProviderCapabilityHTMLHandlerOptions, VoiceProviderCapabilityKind, VoiceProviderCapabilityOptions, VoiceProviderCapabilityReport, VoiceProviderCapabilityRoutesOptions, VoiceProviderCapabilitySummary, } from "./core/providerCapabilities";

package/dist/index.js

@@ -45566,6 +45566,7 @@ var createGeminiVoiceAssistantModel = (options) => {
 var DEFAULT_MAX_ENTRIES = 32;
 var createCachedTTS = (inner, options) => {
   const maxEntries = options.maxEntries ?? DEFAULT_MAX_ENTRIES;
+  const { store } = options;
   const cache = new Map;
   const remember = (key, events) => {
     cache.delete(key);
@@ -45578,6 +45579,18 @@ var createCachedTTS = (inner, options) => {
       cache.delete(oldest);
     }
   };
+  const loadFromStore = async (key) => {
+    if (!store)
+      return null;
+    try {
+      const events = await store.get(key);
+      if (events && events.length > 0) {
+        remember(key, events);
+        return events;
+      }
+    } catch {}
+    return null;
+  };
   return {
     kind: "tts",
     open: async (openOptions) => {
@@ -45608,9 +45621,8 @@ var createCachedTTS = (inner, options) => {
             await session.send(text);
             return;
           }
-          const cached = cache.get(key);
-          if (cached) {
-            for (const event of cached) {
+          const replayEvents = async (events) => {
+            for (const event of events) {
               const replay = {
                 ...event,
                 receivedAt: Date.now()
@@ -45619,12 +45631,22 @@ var createCachedTTS = (inner, options) => {
                 await Promise.resolve(handler(replay));
               }
             }
+          };
+          const cached = cache.get(key) ?? await loadFromStore(key);
+          if (cached) {
+            await replayEvents(cached);
             return;
           }
           capture = [];
           await session.send(text);
-          remember(key, capture);
+          const rendered = capture;
+          remember(key, rendered);
           capture = null;
+          if (store) {
+            try {
+              await store.set(key, rendered);
+            } catch {}
+          }
         }
       };
     }

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);

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@absolutejs/voice",
-	"version": "0.0.22-beta.574",
+	"version": "0.0.22-beta.576",
 	"description": "Voice primitives and Elysia plugin for AbsoluteJS",
 	"repository": {
 		"type": "git",