@absolutejs/voice 0.0.22-beta.478 → 0.0.22-beta.479

package/dist/amdDetector.d.ts

@@ -0,0 +1,25 @@
+import type { Transcript, VoiceSessionHandle, VoiceSessionRecord } from "./types";
+export type VoiceAMDDetectorInput<TContext = unknown, TSession extends VoiceSessionRecord = VoiceSessionRecord, TResult = unknown> = {
+    api: VoiceSessionHandle<TContext, TSession, TResult>;
+    audioLevel: number | undefined;
+    elapsedSinceFirstAudioMs: number;
+    elapsedSinceLastTurnCommitMs: number;
+    partialTranscript: string;
+    session: TSession;
+    transcripts: Transcript[];
+};
+export type VoiceAMDVerdict = {
+    metadata?: Record<string, unknown>;
+    reason?: string;
+};
+export type VoiceAMDDetector<TContext = unknown, TSession extends VoiceSessionRecord = VoiceSessionRecord, TResult = unknown> = {
+    evaluate: (input: VoiceAMDDetectorInput<TContext, TSession, TResult>) => Promise<VoiceAMDVerdict | undefined> | VoiceAMDVerdict | undefined;
+    intervalMs?: number;
+};
+export type MonologueAMDDetectorOptions = {
+    intervalMs?: number;
+    minMonologueMs?: number;
+    reason?: string;
+    requireFirstAudio?: boolean;
+};
+export declare const createMonologueAMDDetector: <TContext = unknown, TSession extends VoiceSessionRecord = VoiceSessionRecord, TResult = unknown>(options?: MonologueAMDDetectorOptions) => VoiceAMDDetector<TContext, TSession, TResult>;

package/dist/index.d.ts

@@ -71,6 +71,8 @@ export { createVoiceSessionListRoutes, createVoiceSessionReplayHTMLHandler, crea
 export { createVoiceAgent, createVoiceAgentSquad, createVoiceAgentTool, } from "./agent";
 export { createAIVoiceModel } from "./aiVoiceModel";
 export type { CreateAIVoiceModelOptions } from "./aiVoiceModel";
+export { createMonologueAMDDetector } from "./amdDetector";
+export type { MonologueAMDDetectorOptions, VoiceAMDDetector, VoiceAMDDetectorInput, VoiceAMDVerdict, } from "./amdDetector";
 export { createVoiceRAGTool } from "./ragTool";
 export type { VoiceRAGCollectionLike, VoiceRAGQueryResult, VoiceRAGSearchInput, VoiceRAGToolArgs, VoiceRAGToolOptions, VoiceRAGToolResult, } from "./ragTool";
 export { createVoiceApiRequestTool, createVoiceDTMFTool, createVoiceEndCallTool, createVoiceTransferCallTool, createVoiceVoicemailDetectionTool, } from "./agentTools";

package/dist/index.js

@@ -3682,6 +3682,66 @@ var createVoiceSession = (options) => {
   const currentTurnAudio = [];
   let fallbackAttemptsForCurrentTurn = 0;
   let fallbackReplayAudioMsForCurrentTurn = 0;
+  const amdDetector = options.amd;
+  let amdEvaluationTimer = null;
+  let amdFired = false;
+  let amdFirstAudioAt;
+  let amdLastTurnCommitAt;
+  let amdLastAudioLevel;
+  const clearAmdEvaluationTimer = () => {
+    if (amdEvaluationTimer) {
+      clearInterval(amdEvaluationTimer);
+      amdEvaluationTimer = null;
+    }
+  };
+  const evaluateAmd = async () => {
+    if (!amdDetector || amdFired) {
+      return;
+    }
+    let snapshot;
+    try {
+      snapshot = await readSession();
+    } catch {
+      return;
+    }
+    const now = Date.now();
+    const verdict = await Promise.resolve(amdDetector.evaluate({
+      api,
+      audioLevel: amdLastAudioLevel,
+      elapsedSinceFirstAudioMs: amdFirstAudioAt === undefined ? 0 : now - amdFirstAudioAt,
+      elapsedSinceLastTurnCommitMs: amdLastTurnCommitAt === undefined ? 0 : now - amdLastTurnCommitAt,
+      partialTranscript: snapshot.currentTurn.partialText,
+      session: snapshot,
+      transcripts: [
+        ...snapshot.transcripts,
+        ...snapshot.currentTurn.transcripts
+      ]
+    }));
+    if (!verdict || amdFired) {
+      return;
+    }
+    amdFired = true;
+    clearAmdEvaluationTimer();
+    try {
+      await api.markVoicemail({
+        metadata: verdict.metadata
+      });
+    } catch (error) {
+      logger.warn("voice amd markVoicemail failed", {
+        error: toError(error).message,
+        sessionId: options.id
+      });
+    }
+  };
+  const startAmdEvaluationTimer = () => {
+    if (!amdDetector || amdEvaluationTimer || amdFired) {
+      return;
+    }
+    const intervalMs = amdDetector.intervalMs ?? 1000;
+    amdEvaluationTimer = setInterval(() => {
+      evaluateAmd();
+    }, intervalMs);
+  };
   const callSilenceTimeoutMs = options.callSilenceTimeoutMs && options.callSilenceTimeoutMs > 0 ? options.callSilenceTimeoutMs : undefined;
   let callSilenceWatchdog = null;
   let callSilenceFired = false;
@@ -4089,6 +4149,8 @@ var createVoiceSession = (options) => {
       recoverable: false,
       type: "error"
     });
+    clearCallSilenceWatchdog();
+    clearAmdEvaluationTimer();
     await closeTTSSession("failed");
     await closeAdapter("failed");
     await persistRecordings();
@@ -4158,6 +4220,8 @@ var createVoiceSession = (options) => {
       sessionId: options.id,
       type: "complete"
     });
+    clearCallSilenceWatchdog();
+    clearAmdEvaluationTimer();
     await closeTTSSession("complete");
     await closeAdapter("complete");
     await persistRecordings();
@@ -5015,6 +5079,7 @@ var createVoiceSession = (options) => {
   };
   const commitTurnInternal = async (reason = "manual") => {
     clearSilenceTimer();
+    amdLastTurnCommitAt = Date.now();
     const session = await readSession();
     if (session.status === "completed" || session.status === "failed") {
       return;
@@ -5262,6 +5327,7 @@ var createVoiceSession = (options) => {
     await ensureAdapter();
     warmTTSSession();
     kickCallSilenceWatchdog();
+    startAmdEvaluationTimer();
   };
   const disconnectInternal = async (event) => {
     clearSilenceTimer();
@@ -5309,7 +5375,11 @@ var createVoiceSession = (options) => {
       const userBytes = conditionedAudio instanceof Uint8Array ? conditionedAudio : conditionedAudio instanceof ArrayBuffer ? new Uint8Array(conditionedAudio) : new Uint8Array(conditionedAudio.buffer, conditionedAudio.byteOffset, conditionedAudio.byteLength);
       captureRecordingChunk("user", userBytes, recordingConfig.userInputFormat);
     }
+    amdLastAudioLevel = audioLevel;
     if (audioLevel >= turnDetection.speechThreshold) {
+      if (amdFirstAudioAt === undefined) {
+        amdFirstAudioAt = Date.now();
+      }
       if (!speechDetected && activeTTSTurnId !== undefined) {
         cancelActiveTTS("barge-in");
       }
@@ -5342,6 +5412,7 @@ var createVoiceSession = (options) => {
     });
     clearSilenceTimer();
     clearCallSilenceWatchdog();
+    clearAmdEvaluationTimer();
     await closeTTSSession(reason);
     await closeAdapter(reason);
     await persistRecordings();
@@ -34864,6 +34935,36 @@ var createAIVoiceModel = (options) => ({
     return output;
   }
 });
+// src/amdDetector.ts
+var createMonologueAMDDetector = (options = {}) => {
+  const minMonologueMs = options.minMonologueMs ?? 8000;
+  const reason = options.reason ?? "monologue-suspected-voicemail";
+  const requireFirstAudio = options.requireFirstAudio ?? true;
+  return {
+    evaluate: ({
+      elapsedSinceFirstAudioMs,
+      elapsedSinceLastTurnCommitMs,
+      session
+    }) => {
+      if (requireFirstAudio && elapsedSinceFirstAudioMs <= 0) {
+        return;
+      }
+      const noTurnsYet = session.turns.length === 0;
+      const monologueElapsed = noTurnsYet ? elapsedSinceFirstAudioMs : elapsedSinceLastTurnCommitMs;
+      if (monologueElapsed < minMonologueMs) {
+        return;
+      }
+      return {
+        metadata: {
+          detector: "monologue",
+          monologueMs: monologueElapsed
+        },
+        reason
+      };
+    },
+    intervalMs: options.intervalMs ?? 1000
+  };
+};
 // src/ragTool.ts
 var DEFAULT_TOOL_NAME = "searchKnowledgeBase";
 var DEFAULT_DESCRIPTION = "Search the knowledge base and return short grounded citations. Use this whenever the caller asks a question that may be answered by indexed reference material.";
@@ -46032,6 +46133,7 @@ export {
   createPhraseHintCorrectionHandler,
   createOpenAIVoiceTTS,
   createOpenAIVoiceAssistantModel,
+  createMonologueAMDDetector,
   createMemoryVoiceTelnyxWebhookEventStore,
   createMemoryVoiceTelephonyWebhookIdempotencyStore,
   createMemoryVoicePlivoWebhookNonceStore,

package/dist/testing/index.js

@@ -5650,6 +5650,66 @@ var createVoiceSession = (options) => {
   const currentTurnAudio = [];
   let fallbackAttemptsForCurrentTurn = 0;
   let fallbackReplayAudioMsForCurrentTurn = 0;
+  const amdDetector = options.amd;
+  let amdEvaluationTimer = null;
+  let amdFired = false;
+  let amdFirstAudioAt;
+  let amdLastTurnCommitAt;
+  let amdLastAudioLevel;
+  const clearAmdEvaluationTimer = () => {
+    if (amdEvaluationTimer) {
+      clearInterval(amdEvaluationTimer);
+      amdEvaluationTimer = null;
+    }
+  };
+  const evaluateAmd = async () => {
+    if (!amdDetector || amdFired) {
+      return;
+    }
+    let snapshot;
+    try {
+      snapshot = await readSession();
+    } catch {
+      return;
+    }
+    const now = Date.now();
+    const verdict = await Promise.resolve(amdDetector.evaluate({
+      api,
+      audioLevel: amdLastAudioLevel,
+      elapsedSinceFirstAudioMs: amdFirstAudioAt === undefined ? 0 : now - amdFirstAudioAt,
+      elapsedSinceLastTurnCommitMs: amdLastTurnCommitAt === undefined ? 0 : now - amdLastTurnCommitAt,
+      partialTranscript: snapshot.currentTurn.partialText,
+      session: snapshot,
+      transcripts: [
+        ...snapshot.transcripts,
+        ...snapshot.currentTurn.transcripts
+      ]
+    }));
+    if (!verdict || amdFired) {
+      return;
+    }
+    amdFired = true;
+    clearAmdEvaluationTimer();
+    try {
+      await api.markVoicemail({
+        metadata: verdict.metadata
+      });
+    } catch (error) {
+      logger.warn("voice amd markVoicemail failed", {
+        error: toError(error).message,
+        sessionId: options.id
+      });
+    }
+  };
+  const startAmdEvaluationTimer = () => {
+    if (!amdDetector || amdEvaluationTimer || amdFired) {
+      return;
+    }
+    const intervalMs = amdDetector.intervalMs ?? 1000;
+    amdEvaluationTimer = setInterval(() => {
+      evaluateAmd();
+    }, intervalMs);
+  };
   const callSilenceTimeoutMs = options.callSilenceTimeoutMs && options.callSilenceTimeoutMs > 0 ? options.callSilenceTimeoutMs : undefined;
   let callSilenceWatchdog = null;
   let callSilenceFired = false;
@@ -6057,6 +6117,8 @@ var createVoiceSession = (options) => {
       recoverable: false,
       type: "error"
     });
+    clearCallSilenceWatchdog();
+    clearAmdEvaluationTimer();
     await closeTTSSession("failed");
     await closeAdapter("failed");
     await persistRecordings();
@@ -6126,6 +6188,8 @@ var createVoiceSession = (options) => {
       sessionId: options.id,
       type: "complete"
     });
+    clearCallSilenceWatchdog();
+    clearAmdEvaluationTimer();
     await closeTTSSession("complete");
     await closeAdapter("complete");
     await persistRecordings();
@@ -6983,6 +7047,7 @@ var createVoiceSession = (options) => {
   };
   const commitTurnInternal = async (reason = "manual") => {
     clearSilenceTimer();
+    amdLastTurnCommitAt = Date.now();
     const session = await readSession();
     if (session.status === "completed" || session.status === "failed") {
       return;
@@ -7230,6 +7295,7 @@ var createVoiceSession = (options) => {
     await ensureAdapter();
     warmTTSSession();
     kickCallSilenceWatchdog();
+    startAmdEvaluationTimer();
   };
   const disconnectInternal = async (event) => {
     clearSilenceTimer();
@@ -7277,7 +7343,11 @@ var createVoiceSession = (options) => {
       const userBytes = conditionedAudio instanceof Uint8Array ? conditionedAudio : conditionedAudio instanceof ArrayBuffer ? new Uint8Array(conditionedAudio) : new Uint8Array(conditionedAudio.buffer, conditionedAudio.byteOffset, conditionedAudio.byteLength);
       captureRecordingChunk("user", userBytes, recordingConfig.userInputFormat);
     }
+    amdLastAudioLevel = audioLevel;
     if (audioLevel >= turnDetection.speechThreshold) {
+      if (amdFirstAudioAt === undefined) {
+        amdFirstAudioAt = Date.now();
+      }
       if (!speechDetected && activeTTSTurnId !== undefined) {
         cancelActiveTTS("barge-in");
       }
@@ -7310,6 +7380,7 @@ var createVoiceSession = (options) => {
     });
     clearSilenceTimer();
     clearCallSilenceWatchdog();
+    clearAmdEvaluationTimer();
     await closeTTSSession(reason);
     await closeAdapter(reason);
     await persistRecordings();

package/dist/types.d.ts

@@ -725,6 +725,7 @@ export type CreateVoiceSessionOptions<TContext = unknown, TSession extends Voice
     trace?: VoiceTraceEventStore;
     recording?: VoiceSessionRecordingConfig;
     callSilenceTimeoutMs?: number;
+    amd?: import("./amdDetector").VoiceAMDDetector<TContext, TSession, TResult>;
     reconnect: Required<VoiceReconnectConfig>;
     phraseHints?: VoicePhraseHint[];
     sessionMetadata?: Record<string, unknown>;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@absolutejs/voice",
-  "version": "0.0.22-beta.478",
+  "version": "0.0.22-beta.479",
   "description": "Voice primitives and Elysia plugin for AbsoluteJS",
   "repository": {
     "type": "git",